Subclonal cooperating lesions lost at relapse - what was actually measured, and why 'lost' is the wrong word
also called subclonal RAS mutations lost at relapse · cooperating lesions are not maintained · inverted clonal architecture in infant KMT2A-r ALL
In most cancers the extra mutations that help the tumour grow are in every cell and are still there when the disease returns. In infant KMT2A leukaemia they are in a minority of cells at diagnosis and often missing when the disease comes back - the opposite pattern. That is strong evidence that the fusion, not the extra mutations, is the disease; but a more sensitive test has since shown that some of those mutations were never gone, only too rare to see.
The same thing, for a scientist
This entry closes an open forward reference: the id `mechanism-subclonal-cooperating-lesions-lost-at-relapse` was cited by `epi-fusion-positive-newborns-versus-leukemia-risk` and had never been written. It exists to say precisely what was measured, because the phrase is used across this graph as if it were a single established fact and it is an aggregate of four measurements with different instruments and different floors. **The claim's shape.** The KMT2A fusion is present in every leukaemic cell and is stable from diagnosis to relapse. The cooperating kinase/PI3K/RAS lesions are not: 65% (20/31) sit below 30% mutant allele frequency at diagnosis despite 92% mean tumour purity, and in matched pairs they are frequently absent at relapse [E1][E2]. **That is the inverse of the standard clonal architecture of leukaemia**, where the cooperating lesion is clonal and the subclones carry the resistance mutations, and it is the most direct genomic argument in this disease that the fusion is close to sufficient. **The number behind 'frequently'.** Seven matched diagnosis/relapse infant pairs were analysed, five of which carried a pathway mutation at diagnosis. The mutation was **maintained or increased in 2** (INF001, INF65), **lost in 2** (INF033, INF042), and in a fifth (INF073) the PIK3CA MAF fell from 39% to 15% [E1]. So 'frequently lost' rests on **2 of 5 lost outright and 1 of 5 declining** - a real signal from five patients, not a majority finding. **Does it survive a depth-aware reanalysis? Partly, and the failure is instructive.** Duplex sequencing of the same disease shows the relapse clone persisting as a minor subclone from diagnosis, through treatment, and expanding only in the last month of disease, with NRAS G13D/G12S and FLT3 D835H clones at **0.06-0.34% of diagnostic cells** expanding at relapse [E3]. At that floor, 'lost at relapse' means *the dominant diagnostic subclone did not survive*, not *the lesion was eliminated*, and a **different** low-frequency signalling clone can be what returns. The structural claim - these lesions are not clonal and are not required for maintenance - survives. The eliminative reading of 'lost' does not.
How clonality was actually called
This is the part the phrase usually travels without, and it is what decides how much weight the claim can carry.
Instrument. Paired-end WGS on diagnostic leukaemia and matched remission marrow or blood, discovery cohort of 22 infants, mean haploid coverage 39x 1. At 39x a variant at 10% allele fraction is expected on about four reads, so subclonality calls from WGS alone are noisy by construction.
Purity. Mean tumour purity 92%, range 74-100%. This matters because low purity mimics subclonality: a 40% MAF in a 50%-pure sample is clonal. High purity is what licenses reading a low MAF as a minor clone rather than as dilution.
Confirmation. Custom capture plus deep sequencing of all tier 1-3 mutations at an average of 437x in five cases, which is the actual evidentiary basis for the intra-tumour-heterogeneity claim - five patients, not 22 1.
Clustering. For the two cases with full diagnosis/relapse/germline trios, tier-3 SNVs - variants in unique sequence outside genes and regulatory regions, chosen because they are unlikely to be under selection - were assigned to statistically significant MAF clusters, then adjusted for purity. Median target coverage in those trios was 1,534x and 894x at diagnosis and 769x and 882x at relapse 1.
The worked case. INF002 gave five clusters at MAF 0.426, 0.236, 0.114, 0.058 and 0.024, i.e. five related clones at 6.25-50% population frequency after purity adjustment. At relapse the majority of tier-3 mutations sat at MAF about 0.5: a single clone, descended from a subclone that had been 6.25% of the diagnostic sample, plus newly acquired mutations. INF001 followed the same pattern 1.
So the clonal-architecture claim is: high-depth, purity-adjusted, cluster-based, and built on two patients for the trajectory and five for the heterogeneity.
Four measurements, four floors
| study | design | depth / floor | what it can say about 'lost at relapse' |
|---|---|---|---|
| St. Jude PCGP 1 | 22 WGS + 43 panel; 7 matched dx/relapse pairs | 39x WGS; 437x capture in 5 cases; 769-1,534x in 2 trios | 2 of 5 pathway mutations lost, 1 declined, 2 retained |
| Interfant-99/06 2 | 124 infant B-ALL, genome-wide | not stated in the abstract | recurrent KRAS/NRAS mutations subclonal and frequently lost at relapse |
| Lund duplex 3 | 20 infants, 7 children; 5 longitudinal, 6 paired relapse | duplex, ~0.01-0.1% | 62% of cancer-associated mutations below standard NGS; relapse clone persists as a minor subclone throughout |
| Lund relapse genomics 4 | 36 relapsed ALL/AML WGS/WES; 257 longitudinal samples in 30 patients | deep targeted longitudinal | relapse mechanism differs by timing: drug-response gene alterations in 79% of early relapse, 9% of very early relapse |
The four are not in conflict. They are four different questions, and the phrase 'subclonal cooperating lesions lost at relapse' has been used as though it answered all of them.
The timing split, and why it makes 'cooperating lesion' ambiguous
The most recent relapse genomics in this disease separates relapses by when they happen 4:
- Early relapse ALL (9-36 months from diagnosis): somatic alterations in drug-response genes, most commonly TP53 and IKZF1, in 79% of cases, with a marked chemotherapy-exposure mutational signature.
- Very early relapse ALL (< 9 months): the same drug-response alterations in 9%, and no chemotherapy signature.
- AML relapse: no chemotherapy signature either; the authors conclude KMT2A-r ALL and AML evade therapy by different routes.
This splits the word 'cooperating' into two populations that should not share an entry. The RAS-pathway lesions present at diagnosis and lost at relapse are one thing. The TP53 and IKZF1 alterations that appear at early relapse carrying a chemotherapy signature are another thing entirely - they are consequences of treatment, not cooperating lesions of the fusion, and they are the ones that persist because they are what was selected.
The graph should keep these separate, and the gene entries in this beat (gene-tp53, gene-ikzf1) are written on that basis.
What this licenses and what it does not
Licensed. The fusion, not the co-mutation, is what maintains this leukaemia. A lesion that the disease discards under therapy in a substantial fraction of patients is not a maintenance requirement, and the expression data make the passenger reading uncomfortable at the same time - every activating pathway mutant allele was expressed irrespective of allele frequency 1.
Not licensed. That these lesions are absent at relapse. At duplex depth they are not absent; the composition changes 3. Any MRD or relapse-prediction argument built on 'the RAS mutation is gone' is built on a detection floor.
Also not licensed. That subclonality means unimportance. The clone that produced relapse in INF002 was 6.25% of the diagnostic sample 1, and the clone that produced resistance in the duplex series was 0.06-0.34% 3. In this disease the clinically decisive clone has repeatedly been a small one.
| closes forward reference | mechanism-subclonal-cooperating-lesions-lost-at-relapse, cited by epi-fusion-positive-newborns-versus-leukemia-risk and unwritten until this cycle |
|---|
measurement table
[
{
"study_pmid": "25730765",
"design": "22 infant MLL-R ALL WGS + 43-case 232-gene panel; 7 matched dx/relapse pairs",
"depth": "39x mean haploid WGS; 437x custom capture in 5 cases; 769-1534x median target coverage in 2 dx/relapse/germline trios",
"purity": "mean 92%, range 74-100%",
"clonality_method": "MAF clustering of tier-3 SNVs outside genes and regulatory regions, purity-adjusted",
"relapse_result": "of 5 informative pairs: 2 maintained or increased, 2 lost, 1 declined 39% to 15% MAF"
},
{
"study_pmid": "30679323",
"design": "124 infant B-ALL, Interfant-99/06, genome-wide multi-layer",
"depth": "not stated in abstract",
"purity": "not stated in abstract",
"clonality_method": "not stated in abstract",
"relapse_result": "recurrent KRAS/NRAS mutations subclonal and frequently lost at relapse"
},
{
"study_pmid": "36204688",
"design": "20 infants, 7 children KMT2A-r ALL; 5 longitudinal, 6 paired relapse",
"depth": "targeted duplex sequencing, floor roughly 0.01-0.1% cellular prevalence",
"purity": "not applicable at this floor",
"clonality_method": "cellular prevalence from duplex consensus reads",
"relapse_result": "NRAS G13D/G12S and FLT3 D835H clones at 0.06-0.34% of diagnostic cells expanded at relapse; relapse clone persisted throughout treatment"
},
{
"study_pmid": "41062506",
"design": "36 relapsed infant/childhood KMT2A-r ALL and AML WGS/WES; 257 longitudinal samples in 30 patients",
"depth": "longitudinal deep targeted sequencing",
"purity": "not stated in abstract",
"clonality_method": "longitudinal tracking of residual leukaemia and clonal drug response",
"relapse_result": "drug-response gene alterations (TP53, IKZF1) in 79% of early relapse ALL vs 9% of very early relapse; chemotherapy-exposure signature only in early relapse ALL"
}
]verdict
{
"survives": "the structural claim - cooperating kinase/PI3K/RAS lesions are subclonal at diagnosis and are not required for maintenance of the leukaemia",
"does_not_survive": "the eliminative reading of 'lost' - at duplex depth the lesions are not absent at relapse, and a different sub-1% signalling clone can be what returns",
"newly_separated": "TP53 and IKZF1 alterations at early relapse carry a chemotherapy-exposure signature and are treatment consequences, not cooperating lesions of the fusion"
}- If the relapse clone is present at 0.06% at diagnosis and persists through treatment, then any therapy that does not reach it merely changes which clone relapses - and a targeted agent chosen against the dominant clone is the most efficient way to select the minor one.
- MRD monitoring keyed to the fusion is reliable because the fusion is clonal; MRD keyed to a cooperating mutation is not, because the cooperating mutation may not be in the relapsing clone. Any assay design that treats a RAS mutation as a trackable marker in this disease inherits that failure.
- The finding that cooperating lesions are dispensable is usually read as good news for menin inhibition - one target, no second driver. It cuts the other way too: with no second lesion to co-target, there is no combination partner available on genomic grounds, and resistance will have to be pre-empted by mechanism rather than by genotype.
- A claim that a lesion was 'lost at relapse' is a claim about a detection floor. This project should not repeat the phrase without the floor, and it has done so in at least one existing entry.
- How many of the pathway mutations scored 'lost at relapse' by bulk sequencing are present below 1% at relapse? The reanalysis is a duplex-sequencing panel run on stored matched pairs and nobody has published it.
- Is the relapse-initiating subclone identifiable at diagnosis prospectively, or only in hindsight? Every published instance of 'the 6.25% clone came back' was identified by sequencing backwards from the relapse.
- Why are the same pathway lesions clonal in non-infant KMT2A-rearranged leukaemia and subclonal in infants? The difference survives within the same fusion class, so it is about the cell or the developmental window, not about the fusion.
- Does a menin inhibitor change the clonal composition of the RAS-pathway subclones? No study has tracked allele frequencies of cooperating lesions under menin inhibition in this disease.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In infant MLL-rearranged ALL, 65% (20/31) of activating kinase/PI3K/RAS mutations had mutant allele frequency below 30% despite 92% mean tumour purity; in 6 of 8 cases with two or more pathway mutations each was below 30%; in 7 matched diagnosis/relapse pairs, of the 5 carrying a pathway mutation at diagnosis it was maintained or increased in 2, lost in 2, and declined from 39% to 15% MAF in a fifth; and clonal reconstruction in case INF002 identified five diagnostic clones at 6.25-50% population frequency with relapse arising from the 6.25% subclone.
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
The relapse statement rests on 5 informative patients; the clonal trajectory on 2. WGS was at 39x mean haploid coverage, so subclonal calls from WGS alone are imprecise, and the deep-capture confirmation covered 5 cases at an average of 437x. Clusters were built on tier-3 SNVs outside genes and regulatory regions, which assumes those variants are not under selection - reasonable but not verified. Full text read via efetch db=pmc (PMC4553269) after Europe PMC fullTextXML returned HTTP 500.
Evidence that argues against this (1)
- Duplex sequencing of the same disease finds relapse clones persisting as minor subclones from diagnosis at 0.06-0.34% of cells, so a pathway mutation scored as lost at 437x may be present two to three orders of magnitude lower. Pilheden M et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An independent series of 124 infant B-ALL cases treated on Interfant-99/06 found recurrent mutations exclusively in K-RAS and N-RAS, subclonal, and frequently lost at relapse.
Unraveling the cellular origin and clinical prognostic markers of infant B-cell acute lymphoblastic leukemia using genome-wide analysis. · Agraz-Doblas A, Bueno C, Bashford-Rogers R, et al. · Haematologica 2019What this evidence cannot show
The abstract does not give the number of matched relapse pairs behind 'frequently lost at relapse', nor the sequencing depth, so this entry treats it as an independent replication of the direction of the finding and not as a second quantitative estimate. Abstract read; full text not retrieved for this entry.
Evidence that argues against this (1)
- The same abstract reports a larger number of non-recurrent, non-silent mutations alongside the recurrent RAS ones, so 'exclusively' describes which genes recurred rather than the full mutational content, and a reader taking it as the latter would understate the genome's complexity. Agraz-Doblas A et al. 2019
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Targeted duplex sequencing of 20 infants and 7 children with KMT2A-rearranged ALL, including 5 longitudinal and 6 paired relapse sample sets, found 62% of cancer-associated mutations below standard NGS resolution, drug-resistant clones carrying NRAS G13D or G12S or FLT3 D835H at 0.06-0.34% of diagnostic cells expanding at relapse, and relapse clones persisting as minor subclones from diagnosis and through treatment before expanding in the last month of disease.
Duplex Sequencing Uncovers Recurrent Low-frequency Cancer-associated Mutations in Infant and Childhood KMT2A-rearranged Acute Leukemia. · Pilheden M, Ahlgren L, Hyrenius-Wittsten A, et al. · HemaSphere 2022What this evidence cannot show
27 patients, of whom only 6 contributed paired relapse material, so the depth advantage comes with a smaller n than the cohorts it qualifies. A targeted panel bounds which genes could be seen. A variant at 0.1% cellular prevalence cannot be assigned to a leukaemic rather than a normal haematopoietic cell by sequencing alone; the inference rests on its expansion at relapse. Abstract read; full text not retrieved for this entry.
Evidence that argues against this (1)
- The same paper reports infants having fewer cancer-associated mutations than the older children, so the low-mutation-burden phenotype of infant disease is not an artefact of shallow sequencing - the deeper instrument reproduced the age effect while lowering the floor. Pilheden M et al. 2022
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Whole-genome and exome sequencing of 36 infants and children with relapsed KMT2A-rearranged ALL or AML, with longitudinal deep sequencing of 257 samples in 30 patients, found somatic alterations in drug-response genes - most commonly TP53 and IKZF1, 64% - highly enriched in early relapse ALL (79%, 9-36 months after diagnosis) but rare in very early relapse ALL (9%, under 9 months), with a marked chemotherapy-exposure mutational signature in early relapse ALL and not in very early ALL or in AML relapse.
The genomic landscape of relapsed infant and childhood KMT2A-rearranged acute leukemia. · Ahlgren L, Pilheden M, Sturesson H, et al. · Nature Communications 2025What this evidence cannot show
A relapse cohort, so it cannot estimate the frequency of anything at diagnosis, and the timing subgroups are small fractions of 36 patients. 'Drug-response genes' is a curated category defined by the authors; the 64% figure is for that category, not for TP53 or IKZF1 individually. Mutational-signature attribution in small mutation sets is unstable. Abstract read; full text not retrieved for this entry.
Evidence that argues against this (1)
- A chemotherapy-exposure signature establishes that the mutations arose under treatment; it does not establish that they caused the relapse, and the very-early relapses - which have neither the signature nor the alterations - are the ones with the worst prognosis, so the mechanism with the clearest genomic footprint is not the mechanism with the worst outcome. Ahlgren L et al. 2025
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In the same cohort, all activating kinase/PI3K/RAS mutant alleles were expressed at RNA level irrespective of their mutant allele frequency, against 38-48% of the other non-silent mutations, and gene-set enrichment within the MLL-AFF1 cases showed expression signatures consistent with RAS-pathway activation.
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
Expression was assessed only in cases with RNA-seq available, a subset of the cohort. Bulk RNA-seq detects that a mutant allele is transcribed somewhere in the sample; it cannot show that it is transcribed in the subclone that carries it, though for a subclonal variant detected in bulk RNA that inference is close to forced. The GSEA signature is a whole-sample readout and a subclonal lesion producing a whole-sample signature would require either paracrine effect or a larger clone than the DNA suggests - which the paper does not resolve.
Evidence that argues against this (1)
- If the RAS-activation expression signature is detectable in bulk RNA from samples where the RAS mutation sits in under 30% of cells, either the signature is not driven by the mutation or the mutant clone is larger than the DNA allele fraction indicates. Neither possibility is addressed, and both weaken the use of the signature as confirmation. Andersson AK et al. 2015
TP53 alterations at early relapse carry a chemotherapy signature and are treatment consequences, not cooperating lesions
the relapse half of the timing split this mechanism entry separates out
states what was measured behind the claim, with the depths and the detection floors
the clonality and relapse evidence the pathway entry summarises, with the number behind 'frequently'
the inverse half: the fusion is clonal and stable while its cooperating lesions are subclonal and unstable
- supports →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the genomic argument that the fusion is close to sufficient: its partners are dispensable at relapse
the same instrument audit from the burden side rather than the clonality side
whether the clonality literature in this disease has a sampling-compartment artefact at its centre