Kinase/PI3K/RAS pathway activation — the only common co-lesion class in KMT2A-r leukemia, and it is mostly subclonal

Two things stand out. First, the frequency is not KMT2A-specific: 47% in KMT2A-r infants versus 39% in non-KMT2A-r infants is not a significant difference (P = 0.59) 1. RAS activation is a feature of infant ALL, not a feature the fusion selects for uniquely — although within infant ALL it does concentrate in t(4;11) 2, and NRAS specifically is enriched in t(4;11) (32% vs 6%, P = 0.01) 3.

Second, the lower Driessen figure is a method effect, not a contradiction. That study used conventional Sanger sequencing of NRAS, KRAS and BRAF only 2; the PCGP figure covers the whole tyrosine-kinase/PI3K/RAS class (adding FLT3, PTPN11, NF1, PIK3CA, PIK3R1, CBL) by whole-genome plus 232-gene targeted sequencing, and Sanger cannot see subclones below roughly 20% allele frequency — which is where two thirds of these mutations live.

The individual genes have their own nodes: comut-kras-kmt2a-r, comut-nras-kmt2a-r, comut-ptpn11-kmt2a-r, comut-nf1-kmt2a-r, comut-flt3-kmt2a-r, comut-pik3r1-kmt2a-r.

This is the most therapeutically consequential number in the beat.

• 65% (20 of 31) of the activating kinase/PI3K/RAS mutations in the PCGP infant cohort sat at a mutant allele frequency below 30%, against an average tumour purity of 92% 1. These are minor diagnostic subclones, not clonal drivers.
• Eight cases carried two or more mutations in the pathway; in six of those eight, every one of the mutations was below 30% MAF 1. That is the signature of parallel, independent, convergent acquisition in separate subclones — several cells independently finding the same pathway — rather than sequential accumulation in one lineage.
• Yet 100% of the activating pathway mutant alleles were expressed at the RNA level, irrespective of allele frequency, against only 48% of non-silent SNVs generally 1. So the subclones are not silent passengers; they are doing something. They are just not doing it in most of the tumour.
• In non-infant KMT2A-r leukemia this inverts: the majority of pathway mutations are in the major clone, and this extends to all somatic exonic mutations (P < 0.0001 versus infants) 1.

The clonal architecture therefore differs by age within the same driver: in an infant, RAS activation is a late subclonal event; in an older child it is closer to a founding event.

For an independent adverse effect. In 109 infant ALL patients, RAS mutation was an independent predictor of poor outcome with a hazard ratio of 3.194 (95% CI 1.211-8.429, P = 0.019). RAS-mutant infants also had higher white-cell counts at diagnosis (P = 0.013) and their blasts were more resistant to glucocorticoids in vitro (P < 0.05) 2. Critically, the same analysis showed that RAS mutation, and not the absence of HOXA9 expression nor the expression of the reciprocal AF4-MLL product, was what associated with poor outcome in t(4;11) infants 2 — a claim that collides directly with the reciprocal-fusion prognostic literature (see cnv-reciprocal-derivative-chromosome-retention).

Against. In the PCGP cohort, patients with an activating pathway mutation showed only a trend toward poorer EFS and OS and increased cumulative incidence of relapse, which did not reach statistical significance, though the trend was stronger when the mutation was present at MAF above 30% 1. The PCGP outcome analysis was limited to 31-33 St. Jude patients, so this is an underpowered non-replication rather than a refutation.

The defensible synthesis: RAS-pathway mutation is probably an adverse marker when it is clonal, and probably close to neutral when it is subclonal, which is most of the time. RE-CHECKED 2026-09-20 and it still holds. A bounded query pairing KMT2A/MLL and infant with the RAS family and allele-frequency/clonality terms and an outcome term returned 8 records, all read, none of which stratifies outcome on RAS allele frequency; controls of the same shape returned 56, 192 and 571. So nobody has tested it by stratifying outcome on allele frequency in a large cohort — which is the obvious, cheap, and still-unrun analysis.

What the re-check did turn up, and this graph had not cited it. PMID 27698462 deep-sequenced RAS in 36 MLL-AF4-positive paediatric patients and found mutations including small subclones in 63.9% - against 25-50% by Sanger - which is a direct measurement of the undercounting this section argues for. In 17 paired samples, relapse RAS clones almost all came from clones already present at diagnosis that survived therapy. Set against that, the 124-patient Interfant-99/06 analysis reports RAS-pathway mutations as subclonal and frequently lost at relapse. Both are true and they pull opposite ways on whether clone size matters - which is precisely why the outcome-by-allele-frequency analysis is still the thing to run.

The cleanest experiment in this area was designed to answer exactly whether KMT2A::AFF1 needs a RAS hit 5.

Human cord-blood CD34+ haematopoietic stem/progenitor cells were transduced with KMT2A::AFF1, KRAS-G12V, or both, and transplanted intra-bone-marrow into immunodeficient mice. Results:

• Clonogenic and differentiation assays: KRAS activation did not cooperate with KMT2A::AFF1 to immortalise CD34+ HSPCs.
• Transplantation: KMT2A::AFF1 alone, KRAS-G12V alone, or both together enhanced haematopoietic repopulation without impairing myeloid-lymphoid differentiation, and without initiating leukemia.
• What KRAS did do: it enhanced extramedullary haematopoiesis of KMT2A::AFF1-expressing cell lines and HSPCs, associated with leukocytosis and CNS infiltration — both hallmarks of infant t(4;11) B-ALL. Transcriptional profiling of KMT2A::AFF1 patient samples supported a cell-migration gene signature underlying the mutant-KRAS phenotype.

The authors' conclusion is that mutant KRAS in KMT2A::AFF1 B-ALL matters for tumour maintenance and dissemination rather than initiation 5. That fits the clinical genetics precisely: a subclonal lesion associated with higher white-cell count 2 and with CNS-tropic behaviour, arriving after the fusion, not with it.

Primary KMT2A-r infant ALL cells were tested ex vivo against three MEK1/2 inhibitors — trametinib, selumetinib and binimetinib (MEK162) 4.

• All RAS-mutant samples were sensitive; the inhibitors severely impaired primary RAS-mutant KMT2A-r infant ALL cells, reduced phospho-ERK and induced apoptosis.
• Among RAS-wild-type cases, both sensitive and resistant samples were found, with no apparent downstream over-activation detectable in the wild-type samples to explain the sensitive ones.
• MEK inhibition synergistically enhanced prednisolone sensitivity in both RAS-mutant and RAS-wild-type cells 4.

That last point is the interesting one, and it is the strongest argument for pursuing MEK inhibition despite the subclonality problem. If the benefit is glucocorticoid sensitisation rather than killing a RAS-mutant subclone, then the subclonal architecture and the loss of RAS mutations at relapse stop being disqualifying — the drug would be acting on the bulk population through a different mechanism, and RAS genotype would be the wrong selection biomarker. This also dovetails with the observation that RAS-mutant blasts are glucocorticoid-resistant in vitro 2: MEK inhibition may be reversing precisely the resistance the co-mutation confers.

No randomised trial of MEK inhibition in infant KMT2A-r ALL has reported. The realistic competitor for the same combination slot is now blinatumomab, which raised 2-year disease-free survival from 49.4% to 81.6% on the Interfant-06 backbone 6 and sets a bar that a MEK inhibitor would have to clear or complement.

The standing rule in this project is one entry per thing; never a summary entry covering several. This entry covers the RAS pathway as a unit, and until cycle 23 that was partly defensible because the members had no entries. They do now: gene-kras, gene-nras, gene-ptpn11, gene-nf1 and comut-pik3r1-kmt2a-r, each edged part_of this entry.

So this entry should be read as a pathway-level overview and not as the record for any member gene. The distinction matters here more than it usually would, because the members do not behave alike: PIK3R1's lesion is an inter-SH2 internal tandem duplication, a class that SNV pipelines handle worst, and it is the only cooperating lesion in this disease with a functional demonstration on the patients' own alleles. A pathway-level summary averages that away.

And the identifier hazard is unusually sharp in this family (brief rule 68, verified against live UniProt on 2026-09-20): KRAS, NRAS and HRAS are all exactly 189 aa, share hotspot codons, and HRAS1 is NRAS's only UniProt synonym while also being HRAS's synonym. The length assertion that verify-accessions.py relies on offers no protection at all here. A source that says only 'RAS mutation' is making a pathway claim, and this entry is where it belongs - it must not be silently assigned to a member gene.