The RAS-mutant subclone is where the leukaemia goes, not what it is - a dissemination modifier that a marrow aspirate systematically under-counts
also called RAS lesions confer extramedullary homing in infant KMT2A-r ALL · the CNS clone hypothesis · why a subclonal RAS mutation is prognostically adverse
In infant KMT2A leukaemia the RAS mutation is usually in a minority of cells, yet it predicts a worse outcome - which does not fit a passenger mutation. The proposal here is that those cells are the ones that leave the bone marrow and spread, including into the brain and spinal fluid, so a bone-marrow sample sees only the fraction that stayed behind.
The same thing, for a scientist
Three observations sit awkwardly together in this disease. RAS-pathway mutations are **subclonal in about two thirds of cases** and frequently absent at relapse [E1][E2], which argues against a driver. They are **expressed in 100% of cases irrespective of allele frequency** against 38-48% of other non-silent mutations [E1], which argues against a passenger. And they are **independently prognostically adverse** (HR 3.194 in Interfant) with higher white cell counts at diagnosis [E3]. The experimental observation that reconciles them is [E4]: KRAS(G12V) added to MLL-AF4 in human cord-blood CD34+ HSPCs **did not initiate leukaemia** but enhanced **extramedullary engraftment, leukocytosis and CNS infiltration**, with a cell-migration transcriptional signature in patient samples. **Hypothesis.** The RAS-mutant subclone in infant KMT2A-rearranged ALL is a **dissemination-competent subpopulation**, not a growth-advantaged one. Its contribution to outcome is through extramedullary and CNS burden rather than through marrow proliferation, and its measured allele fraction in a marrow aspirate is therefore **a systematic underestimate** of its true representation in the patient - because the cells that carry it are disproportionately elsewhere. **If true, four things follow.** (i) RAS allele fraction should be **higher in CSF or extramedullary blasts than in paired marrow blasts** from the same patient at the same time. (ii) 'Lost at relapse' in a marrow-sampled series partly reflects **where relapse was sampled**, not what survived. (iii) The prognostic effect should be **mediated by CNS and extramedullary disease**, and should attenuate when those are adjusted for. (iv) MEK inhibition should act primarily on **compartment distribution** - CNS burden, leukocytosis - rather than on marrow blast count, which is the endpoint every study has measured. **Prior probability: 0.35.** The mechanism has direct experimental support in a human model [E4] and the prognostic and clonality data are consistent with it, but the central prediction - a compartment gradient in allele fraction - has never been tested in a patient, and there is a strong simpler rival (the prognostic effect is driven by the clonal minority of RAS mutations, which is what the St. Jude data hint at when the trend concentrates in cases with MAF above 30% [E1]).
The rival explanation, stated fairly
Rival: only the clonal RAS mutations matter, and the subclonal ones are noise.
Evidence for it. In the St. Jude cohort the trend toward worse EFS and OS was concentrated in cases with MAF above 30% 1. The Interfant hazard ratio was derived from Sanger sequencing, which has a floor around 15-20% VAF and therefore detects mainly clonal mutations - so HR 3.194 may already be a hazard ratio for clonal RAS mutation and say nothing about the subclonal majority 3. And a 488-patient paediatric B-ALL cohort found RAS-pathway alterations as a class not significantly associated with survival 5.
This rival is simpler and requires no new mechanism. It is also distinguishable from the hypothesis by the compartment prediction: the rival predicts no CSF-versus-marrow allele-fraction gradient, the hypothesis predicts one. That is the experiment worth doing.
A third possibility both must be tested against. The NG2 antigen has been implicated in leukaemia invasiveness and CNS infiltration in KMT2A-rearranged infant B-ALL, and is already cited in this graph's comut-kras-kmt2a-r. If CNS tropism in this disease is an NG2 phenomenon, the RAS association could be correlational.
Why this is worth testing even though the lesion is not the driver
CNS involvement is a first-order clinical problem in infant KMT2A-rearranged ALL, and CNS-directed therapy is among the most toxic parts of an infant protocol. A biomarker that identified which patients' leukaemia is compartment-mobile would be useful regardless of whether the lesion drives proliferation - which is precisely the kind of usefulness a driver-versus-passenger framing makes invisible.
It also reframes a therapeutic question. If MEK inhibition acts on dissemination rather than on bulk, then the endpoint on which MEK inhibitors have been evaluated in this disease - ex-vivo cytotoxicity, marrow burden - is the wrong one, and a negative result on that endpoint would not have excluded a real effect. The one in-vivo combination test in KMT2A-rearranged RAS-mutant leukaemia reduced burden without extending survival; nobody reported compartment distribution.
| statement | In infant KMT2A-rearranged ALL the RAS-mutant subclone is dissemination-competent rather than growth-advantaged: it is enriched in extramedullary and CNS compartments relative to bone marrow, its prognostic effect is mediated by extramedullary and CNS burden, and its measured marrow allele fraction systematically underestimates its representation in the patient. |
|---|---|
| origin | data-pattern |
| prior probability | 0.35 |
predictions
[ "RAS mutant allele fraction is higher in CSF or extramedullary blasts than in paired bone-marrow blasts from the same patient at the same timepoint", "the prognostic effect of RAS mutation attenuates substantially when CNS status and extramedullary disease are included in the model", "apparent loss of a RAS lesion at relapse is less frequent when relapse is sampled outside the marrow", "MEK inhibition changes compartment distribution - CNS burden, circulating blast count - more than it changes marrow blast count", "single-cell or compartment-resolved data show the RAS lesion and the migration transcriptional signature in the same cells" ]
experiments
[ "exp-csf-versus-marrow-ras-allele-fraction" ]
analyses
[ "ana-cooperating-lesion-enumeration-kmt2a-r" ]
falsifiers
[ "Paired CSF and marrow sampling in RAS-mutant infant KMT2A-rearranged ALL shows equal or lower RAS allele fraction in the CSF compartment across a reasonable series - the hypothesis's central prediction fails and it should be withdrawn.", "The Interfant prognostic effect survives adjustment for CNS status and extramedullary disease with an essentially unchanged hazard ratio - the mediation claim fails.", "Single-cell data show the migration signature distributed independently of RAS genotype, or show it in the RAS-wild-type majority - the mechanism is NG2 or something else, not RAS.", "A prognostic analysis stratified by mutant allele fraction shows the entire effect carried by clonal (MAF > 30%) mutations with no effect in the subclonal group - the simpler rival wins and the subclone needs no special explanation." ]
- If the hypothesis is right, MEK inhibition would be a CNS-directed therapy, and CNS penetration becomes the governing pharmacological property - trametinib and selumetinib have limited CNS exposure, so the drugs available would be aimed at the compartment they reach least.
- A compartment-enriched target is worse than a subclonal one for the same reason: treatment selects the clone that stayed in the marrow, which is the clone that MRD monitoring sees and treats as the disease.
- If the prognostic effect is mediated by CNS disease, then reducing the RAS clone without reducing CNS burden would improve a genotype and not an outcome - and genotype is what a trial of a targeted agent would most easily measure.
- The hypothesis, if wrong in the direction of the simpler rival, would have licensed treating subclonal RAS mutations as meaningful when they are not - which is the error this whole beat exists to avoid.
- Has any study in this disease sequenced CSF blasts and marrow blasts from the same patient at the same time? No such report was found in the searches run for this beat.
- Is the migration signature reported in patient samples detectable in the RAS-mutant cells specifically, or is it a bulk property of RAS-mutant cases? Bulk RNA cannot distinguish these and single-cell data in this disease exist.
- If the RAS clone is compartment-mobile, does it also seed the testis and other sanctuary sites that infant protocols do not routinely sample?
- Does the same logic apply to the FLT3 D835H subclones found at 0.06-0.34% by duplex sequencing, or is compartment mobility specific to RAS?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In infant MLL-rearranged ALL, 65% of activating kinase/PI3K/RAS mutations were below 30% mutant allele frequency at 92% mean tumour purity; all such mutant alleles were expressed irrespective of allele frequency against 38-48% of other non-silent mutations; and the trend toward worse event-free and overall survival was present but not significant, concentrated in cases with mutant allele frequency above 30%.
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 survival analysis is on a small cohort and explicitly reported as a trend. Expression was assessed only in cases with RNA-seq. All samples in this study are bone marrow or peripheral blood; no extramedullary or CSF compartment was sequenced, so the hypothesis's central prediction is untested rather than contradicted here. Full text read via efetch db=pmc (PMC4553269).
Evidence that argues against this (1)
- That the survival trend concentrates in cases with MAF above 30% is direct support for the rival explanation - that only clonal RAS mutations carry prognostic weight - and this entry treats it as the leading alternative rather than burying it. Andersson AK et al. 2015
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In 124 infant B-ALL cases from Interfant-99/06, the recurrent K-RAS and N-RAS mutations were 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
Abstract read. The sampling compartment for the relapse samples is not stated in the abstract; in practice relapse sampling in this disease is overwhelmingly bone marrow, which is the confound this hypothesis proposes. Number of matched pairs not given.
Evidence that argues against this (1)
- If relapse samples were bone marrow, 'lost at relapse' is measured in the compartment the hypothesis predicts the RAS clone to be depleted from - so the same observation is evidence for the hypothesis and cannot also be used against it. This entry therefore treats it as neutral rather than supportive. Agraz-Doblas A et al. 2019
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In 109 infant ALL patients, RAS mutation was an independent predictor of poor outcome (HR 3.194, 95% CI 1.211-8.429, P = 0.019) and RAS-mutated infants had significantly higher white blood cell counts at diagnosis (P = 0.013) and greater in-vitro glucocorticoid resistance.
Frequencies and prognostic impact of RAS mutations in MLL-rearranged acute lymphoblastic leukemia in infants. · Driessen EM, van Roon EH, Spijkers-Hagelstein JA, et al. · Haematologica 2013What this evidence cannot show
Sanger sequencing detects mainly clonal mutations, so this hazard ratio probably describes clonal RAS mutation. The higher white cell count is the observation most consistent with the dissemination hypothesis - leukocytosis is the circulating-compartment readout - but leukocytosis is also the classic marker of proliferative burden, so it does not discriminate between the two models on its own. Wide confidence interval, retrospective subgroup.
Evidence that argues against this (1)
- A 488-patient paediatric B-ALL cohort found RAS-pathway alterations as a class not significantly associated with inferior overall or event-free survival, so the prognostic premise of this hypothesis is not reproduced outside infant KMT2A-rearranged disease. Chen X et al. 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KRAS(G12V) did not cooperate with MLL-AF4 to immortalise human cord-blood CD34+ HSPCs or to initiate leukaemia on intrabone marrow transplantation into immunodeficient mice, but enhanced extramedullary haematopoiesis, leukocytosis and central nervous system infiltration, and transcriptional profiling of MLL-AF4-expressing patients supported an underlying cell-migration gene signature.
Activated KRAS Cooperates with MLL-AF4 to Promote Extramedullary Engraftment and Migration of Cord Blood CD34+ HSPC But Is Insufficient to Initiate Leukemia. · Prieto C, Stam RW, Agraz-Doblas A, et al. · Cancer Research 2016What this evidence cannot show
Cord-blood CD34+ HSPCs with lentiviral MLL-AF4, not the fetal-liver target cell this graph favours nor an endogenous-locus model; the extramedullary phenotype is measured in non-leukaemic engrafted cells, so the model shows KRAS changes where MLL-AF4-expressing HSPCs go, not where leukaemic blasts go. Immunodeficient mice have an altered CNS barrier and no adaptive immunity. The patient migration signature is a bulk transcriptomic correlate, not a clone-resolved measurement.
Evidence that argues against this (1)
- NG2 antigen has been independently implicated in leukaemia invasiveness and CNS infiltration in KMT2A-rearranged infant B-ALL, offering a mechanism for CNS tropism that does not require a RAS lesion, so the RAS-CNS association could be correlational. Prieto C et al. 2018
- E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In 488 paediatric B-ALL cases, RAS-pathway alterations as a class were not significantly associated with inferior overall or event-free survival, with the adverse signal confined to FLT3/NF1 co-mutation.
The clinical relevance of RAS pathway gene mutations in childhood B-cell acute lymphoblastic leukemia. · Chen X, Gao L, Zhang Y, et al. · Annals of Hematology 2026What this evidence cannot show
Unselected paediatric B-ALL rather than infant KMT2A-rearranged disease, single centre, retrospective, variants called from whole-transcriptome sequencing. It is cited here as the strongest available challenge to the prognostic premise, not as a measurement in the index population. Abstract read.
Evidence that argues against this (1)
- Because the cohort is unselected paediatric B-ALL, a real effect confined to KMT2A-rearranged infants would be diluted below detection, so this null result constrains the generality of the prognostic claim rather than refuting it in this disease. Driessen EM et al. 2013
built on the KRAS sufficiency and clonality evidence
the compartment prediction, testable in material infant protocols already collect
a mechanism that reconciles subclonal, expressed and prognostically adverse