Fusion partner predicts the molecular subgroup almost perfectly, and survival not at all
also called integrative clusters infant ALL · IC1-5 infant KMT2A · IRX HOXA partner clusters
In 84 babies with this leukaemia, which gene the KMT2A gene is joined to almost completely determines which molecular type of disease they have. It does not determine how long they live.
The same thing, for a scientist
Multi-omics clustering of **84 infants with KMT2A-rearranged leukaemia** produces five integrative clusters, three of which **correlate with the fusion partner at extraordinary significance**: IC3 HOXA-type **MLLT1** (Fisher's exact **P = 3.7 × 10⁻⁸**), IC4 HOXA-type **MLLT3** (P = 5.6 × 10⁻⁶), IC5 HOXA-type **AFF1** (P = 5.7 × 10⁻⁶), alongside IRX-type differentiated (IC1) and undifferentiated (IC2). Yet **no survival differences were observed when all fusion partners were included.** The partner determines the biology and not the outcome, in the index population.
Why this is the sharpest version of the beat's question
The Q19 beat found that KMT2A fusion partners separate prognostically in childhood AML and not in infants — the Moscow cohort gives AFF1 2-year EFS 36% against 37% for other partners, P=0.9. The obvious sceptical reading is that infant cohorts are small and the partner effect is simply undetectable there.
This paper closes that escape. In the same population, with the same modest n, the partner is recoverable from the molecular data at P = 3.7 × 10⁻⁸. The biology is not subtle and the cohort is not underpowered to see partner-linked structure. It is specifically the survival axis on which the partner does not separate.
What does track outcome is the other axis: the IRX/HOXA split, with IC2 infants exhibiting the poorest EFS, and a trend toward the previously reported poor prognosis of the IRX subtype among KMT2A-AFF1 cases. So infant KMT2A-r ALL has prognostic substructure — it is just not indexed by the 3' partner.
How this entry came to exist
It is a bounded rejection that was re-checked and overturned, which is the practice working as intended rather than a failure.
The Q19 enumeration set this paper aside, recording why: it "subgroups by transcriptional state and RAS burden rather than by 3-prime partner". Crucially the agent also wrote the condition for its own reversal: "the paper may contain a partner-by-subgroup table not reflected in its abstract, in which case this rejection is wrong and should be revisited."
scripts/reading-queue.py surfaced the item on low claim-to-abstract overlap, the full text was read, and the condition was met. A rejection that names what would overturn it is recoverable; one that does not is invisible forever. This is the first time in this project that a recorded rejection has been reversed by reading it back.
- Retrieve Supplementary Figures 1e/1f and 5e. The survival comparisons that make this entry's second claim live in them, and without per-arm n and a confidence interval the null is uninterpretable.
- Does the partner separate outcome WITHIN the HOXA subtype? IC3, IC4 and IC5 are MLLT1, MLLT3 and AFF1 respectively; a three-arm comparison restricted to HOXA patients is the partner-by-lineage interaction test the Q19 beat proposed, already half-assembled here.
- Reconcile with `epi-kmt2a-b-all-partner-and-mrd-clearance-across-ages`, which found the partner an independent determinant of MRD clearance across 465 B-ALL cases. If the partner predicts early response but not survival, that gap is itself the finding.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Multi-omics clustering of 84 infants with KMT2A-rearranged leukaemia yields five integrative clusters whose membership correlates strongly with the 3' fusion partner: IC3 HOXA-type MLLT1 (Fisher's exact P = 3.7e-8), IC4 HOXA-type MLLT3 (P = 5.6e-6) and IC5 HOXA-type AFF1 (P = 5.7e-6), with IRX-type divided into IC1 differentiated and IC2 undifferentiated. Other characteristics including age did not correlate significantly with the clusters.
Multi-omics analysis defines highly refractory RAS burdened immature subgroup of infant acute lymphoblastic leukemia · Nature Communications 2022What this evidence cannot show
Read in full from PMC9427775 by the coordinator on 2026-09-19. The discovery cohort is 61 infants with 23 further cases assigned by a k-nearest-neighbours classifier, so the 84 are not a single uniformly analysed set and the cluster labels for the additional 23 are predicted rather than derived. The P values are for correlation between cluster membership and partner, not for any outcome difference.
Evidence that argues against this (1)
- Partner and molecular cluster being near-identical raises the possibility that the clusters add nothing beyond the partner — an expensive re-derivation of the karyotype. The paper's own answer is that the IRX/HOXA axis cuts across partners and carries the prognostic signal, but a reader should not take cluster membership as independent information about a patient whose partner is already known. Nature Communications 2022
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In the same 84-infant cohort, no survival differences were observed when all fusion partners were included, while IC2 infants exhibited the poorest event-free survival and a trend toward the previously reported poor prognosis of the IRX subtype was observed among KMT2A::AFF1 cases.
Multi-omics analysis defines highly refractory RAS burdened immature subgroup of infant acute lymphoblastic leukemia · Nature Communications 2022What this evidence cannot show
Read in full on 2026-09-19. The survival comparisons sit in supplementary figures (Supplementary Fig. 1e, f and 5e) which were not retrieved, so no hazard ratio, confidence interval or per-arm n is recorded here and none should be quoted from this item. 'Trend' and 'poorest' are the authors' words without attached statistics in the main text.
Evidence that argues against this (2)
- A null survival result in 84 infants split across five clusters leaves arms of roughly 15-20 patients, which is underpowered for anything but a large effect. This supports 'no large partner effect' and cannot exclude a moderate one - the same limitation that applies to every infant cohort in the Q19 beat, including the Moscow one it agrees with. International Journal of Laboratory Hematology 2025
- If the prognostic axis is IRX versus HOXA and the partner determines which HOXA cluster a patient falls into, then partner and outcome are not independent - they are related through a variable this analysis holds separately. A null partner effect computed across both IRX and HOXA patients could mask opposing effects within each. Nature Communications 2022