Hypothesis: the cell-of-origin methylome, not the breakpoint, decides whether a KMT2A::AFF1 leukemia is HOXA-high or HOXA-low
also called methylome determines HOXA status · epigenome-of-origin hypothesis
The fusion protein finds its genes by reading which DNA sites are chemically untagged. So the genes it switches on should depend on the chemical state of the cell it landed in — which would mean an infant's fetal cells and an older child's cells give different diseases from the same fusion.
The same thing, for a scientist
The KMT2A CXXC domain reads unmethylated CpG and actively protects specific CpGs from methylation; LEDGF's PWWP reads H3K36me2/3. Fusion targeting is therefore a read-out of the pre-existing epigenome rather than a sequence code. Within t(4;11) ALL, HOXA-low cases are infants with KMT2A breakpoints in introns 11/12 and HOXA-high cases are non-infants with breakpoints in introns 9/10 — but age and breakpoint are confounded in the only primary dataset. We hypothesise the causal variable is the methylome and chromatin state of the fetal versus postnatal cell of origin, and that the breakpoint association is incidental to when the break occurred. Prediction: HOXA status is predictable from methylation data with age controlled, and residual breakpoint effect is small.
The argument, and what separates it from the alternative1 deeper
Three facts, each independently verified elsewhere in this beat.
- Targeting reads the methylome. CXXC point mutants that lose unmethylated-CpG binding lose transactivation and transformation 1, and structure-guided mutants permit specific Hoxa9 CpGs to become methylated, followed by H3K9 methylation, loss of transcription and failure to cause leukemia 2. The domain is not just an address label — it protects particular CpGs.
- The HOXA split exists and co-segregates with age and breakpoint. Infants with low HOXA and breaks in introns 11/12; non-infants with high HOXA and breaks in introns 9/10 3.
- Infant KMT2A-r methylomes are subtype-specific and map onto normal haematopoietic chromatin states. In 69 Interfant-treated patients, methylome signatures map to enhancer and transcriptional-control chromatin states of normal haematopoietic cells and segregate KMT2A-r from non-KMT2A-r 4. And methylation differs sharply by partner — t(4;11)/t(11;19) extensively hypermethylated, t(9;11) resembling normal marrow 5.
The competing hypothesis is breakpoint-driven: intron 11/12 breaks retain more KMT2A sequence, which changes the complex and hence the output. That is perfectly plausible and this hypothesis does not claim it is false — only that it is probably the minor term.
The two are separable because age and breakpoint are confounded in the 2009 dataset but need not be in a modern one. Any cohort with (a) KMT2A breakpoint intron calls, (b) methylation arrays and (c) HOXA-cluster expression allows the effect of methylation on HOXA status to be estimated with breakpoint and age as covariates. Interfant-99/06 and TARGET/St. Jude infant-ALL resources plausibly contain all three.
Why it matters beyond curiosity
If the methylome of origin sets the target set, three things follow that change what the lab does.
Biomarker. HOXA9/MEIS1 transcripts become an unsafe pharmacodynamic readout across the KMT2A-r population, and the readout should instead be fusion-complex eviction or gene-body H3K4me3 (which is menin-inhibitor-sensitive).
Model choice. Every standard model is HOXA-high. A methylome-driven mechanism means the correct model for infant disease is a fetal-derived progenitor, not a transformed line, and it predicts that expressing KMT2A::AFF1 in fetal versus adult CD34+ cells will yield different target sets — an experiment that is feasible today.
Therapeutic caution. Hypomethylating agents would be predicted to broaden the fusion's accessible target set, not narrow it. That is a specific mechanistic reason why azacitidine priming might have failed to help in infants, and it is testable.
| statement | The HOXA-high versus HOXA-low phenotype in KMT2A::AFF1 leukemia is determined primarily by the DNA-methylation and chromatin state of the cell of origin (fetal versus postnatal), with KMT2A breakpoint position a minor contributor whose apparent effect arises from its confounding with age. |
|---|---|
| origin | literature-gap |
| prior probability | 0.35 |
predictions
[ "In a cohort with methylation arrays, breakpoint intron calls and HOXA-cluster expression, methylation features will predict HOXA status with breakpoint and age included as covariates, and the residual breakpoint effect will be small.", "Expressing KMT2A::AFF1 in fetal-liver-derived versus adult-bone-marrow-derived CD34+ cells will produce different fusion target sets and different HOXA output despite an identical construct.", "Unmethylated-CpG-island annotation of the pre-leukemic cell type will predict which loci the fusion occupies better than any sequence motif.", "Treatment with a hypomethylating agent will BROADEN the fusion's accessible target set (more unmethylated CpG islands available), measurable as increased fusion binding-site number by ChIP after azacitidine." ]
experiments
[ "exp-kmt2a-aff1-degron-occupancy-titration" ]
analyses
[ "ana-infant-methylome-hoxa-predictor" ]
falsifiers
[ "Breakpoint intron predicts HOXA status independently of methylation and age in a multivariable model — the hypothesis is then wrong and the breakpoint model is right.", "KMT2A::AFF1 expressed in fetal versus adult CD34+ cells produces the same target set and the same HOXA output.", "HOXA-low cases are found to have normal HOXA-locus methylation, removing the proposed mechanism for their phenotype.", "Hypomethylating-agent treatment does not change the number or identity of fusion binding sites." ]
- If the methylome sets the target set, then hypomethylating agents should broaden rather than narrow the fusion's reach — a mechanistic argument that azacitidine could in principle be harmful, not merely ineffective.
- A methylome-determined target set means the disease is intrinsically heterogeneous and a single pharmacodynamic biomarker cannot serve the whole population.
- If HOXA-low disease runs an unidentified alternative program, all preclinical efficacy data from HOXA-high models over-state expected benefit in infants.
- Does any existing infant-ALL cohort contain methylation arrays, breakpoint intron calls and HOXA expression together? If yes, this hypothesis can be tested computationally this cycle at essentially zero cost.
- Would the same logic explain why t(9;11) infant ALL has a near-normal methylome — i.e. is partner identity also confounded with cell of origin and timing?
- If methylation determines the target set, does that make the pre-leukemic fetal progenitor methylome a prenatal risk biomarker, connecting this beat to the Q1 fusion-genesis work?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Point mutations in the KMT2A CXXC domain that abolish binding to non-methylated CpG DNA also abolish transactivation and myeloid transformation, establishing that fusion targeting depends on reading the DNA-methylation state rather than a sequence motif.
Binding to nonmethylated CpG DNA is essential for target recognition, transactivation, and myeloid transformation by an MLL oncoprotein · Ayton PM, Chen EH, Cleary ML · Molecular and Cellular Biology 24(23):10470-10478 2004What this evidence cannot show
In-vitro myeloid immortalisation; establishes the dependence on methylation state but not that physiological methylome variation changes the target set.
Evidence that argues against this (1)
- Unmethylated CpG is necessary but not sufficient, and the second requirement points at the current transcriptional state rather than at an inherited methylome. Okuda et al. showed that exactly two targeting modules are necessary and sufficient for KMT2A-fusion target recognition - one binding non-methylated CpG and one binding H3K36me2/3 - and that an artificial protein carrying those two modules plus an AF4-coactivator interaction domain functionally substitutes for the fusion. Because H3K36me2/3 is co-transcriptionally deposited, the fusion targets genes that are *already active*, so the target set is set by ongoing transcription as much as by any methylome of origin. Okuda H et al. 2014
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Structure-guided CXXC point mutants in MLL-AF9 permitted specific CpGs at Hoxa9 to become DNA-methylated, followed by increased H3K9 methylation, decreased Hoxa9 transcription, loss of immortalisation and failure to induce leukemia — showing that protection of specific CpGs from methylation is the selective event.
Structure of the MLL CXXC domain-DNA complex and its functional role in MLL-AF9 leukemia · Cierpicki T, Risner LE, Grembecka J, et al. (Zeleznik-Le NJ, Bushweller JH, senior authors) · Nature Structural & Molecular Biology 17(1):62-68 2010What this evidence cannot show
A single locus (Hoxa9) carries most of the mechanistic weight; mouse myeloid model.
Evidence that argues against this (1)
- The CXXC mutant experiments show that protecting CpGs from methylation is required, which is a statement about maintenance, not about a pre-existing imprint. Okuda et al.'s two-module result implies that a cell with the same CpG methylation landscape but different H3K36me2/3 distribution would give a different target set, so the methylome of origin cannot be the sole determinant. Okuda H et al. 2014
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Within pediatric t(4;11) ALL, patients divide into infants with low HOXA expression and KMT2A breakpoints in introns 11/12 versus non-infants with high HOXA expression and breakpoints in introns 9/10.
Two independent gene signatures in pediatric t(4;11) acute lymphoblastic leukemia patients · Trentin L, Giordan M, Dingermann T, Basso G, Te Kronnie G, Marschalek R · European Journal of Haematology 83(5):406-419 2009What this evidence cannot show
Small cohort on 2009 microarrays; age and breakpoint are confounded by design and cannot be separated in this dataset, which is the entire reason this hypothesis is stated rather than asserted as fact. No verified independent replication.
Evidence that argues against this (1)
- Breakpoint position and age are confounded in this split, so the division cannot attribute HOXA output to either. Infants differ from non-infants in cell of origin, in time since the initiating event and in every developmental variable at once; the intron 11/12 versus 9/10 difference travels with age rather than varying independently of it. Oncofusion knowledge graph (internal analysis) 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In 69 uniformly treated infant B-ALL patients, DNA methylome signatures map onto chromatin states related to enhancer and transcriptional control in normal haematopoietic cells and segregate KMT2A-rearranged from non-rearranged cases, with coordinated AP-1 and RUNX enrichment tracking hypomethylated CpGs at those genes.
Integrative methylome-transcriptome analysis unravels cancer cell vulnerabilities in infant MLL-rearranged B cell acute lymphoblastic leukemia · Tejedor JR, Bueno C, Vinyoles M, et al. (Menendez P, senior author) · Journal of Clinical Investigation 131(13):e138833 2021What this evidence cannot show
Does not analyse HOXA-high versus HOXA-low status or KMT2A breakpoint position, so the connection to this hypothesis is ours. Array-based methylome.
Evidence that argues against this (1)
- Methylome-based cell-of-origin inference from bulk leukaemic marrow is confounded by cell composition: Houseman et al. showed DNA methylation signatures function as surrogate measures of leukocyte mixture, so apparent 'origin' signal can be blast fraction and lineage composition. Houseman EA et al. 2012
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Promoter CpG-island methylation patterns separate genetic subtypes of KMT2A-rearranged infant ALL, with t(4;11) and t(11;19) extensively hypermethylated while t(9;11)-positive and KMT2A-germline infant ALL epigenetically resemble normal bone marrow — demonstrating that methylome state varies sharply within the KMT2A-r class.
Specific promoter methylation identifies different subgroups of MLL-rearranged infant acute lymphoblastic leukemia, influences clinical outcome, and provides therapeutic options · Stumpel DJPM, Schneider P, van Roon EHJ, et al. (Stam RW, senior author) · Blood 114(27):5490-5498 2009What this evidence cannot show
Pre-array-era differential methylation hybridisation; the partner-methylation association is itself confounded with age, so it illustrates rather than resolves the confound.
Evidence that argues against this (1)
- The same data support a simpler account than the one this hypothesis needs. If t(9;11) infant ALL is epigenetically like normal marrow while t(4;11) is heavily hypermethylated, the methylome is tracking the fusion subtype rather than setting its target set — which reverses the direction of causation the entry proposes. Oncofusion knowledge graph (internal analysis) 2026
extends the origin-methylome logic from diagnosis to post-treatment destination
Causality counter (Schlesinger 2007, PMID 17200670): CpG islands that acquire de novo methylation in cancer are preferentially those already carrying Polycomb H3K27me3 from the cell of origin, with the EZH2 complex recruiting DNMTs - so 11 hypermethylated miRNA promoters in t(4;11) are compatible with a pre-existing fetal-progenitor repressive state and do not on their own implicate the fusion
would explain why the fetal context matters so much
The variable that separated leukemogenic from non-leukemogenic CXXC swaps was protection of specific CpGs from de novo methylation, not locus targeting — which puts the methylome, not the reader domain, in the causal seat
- related to →KMT2A / MLL1 — the gene that breaks
Counter-evidence (Okuda 2014, PMID 24465000): fusion targeting needs non-methylated CpG AND H3K36me2/3, so target choice tracks genes that are already transcriptionally active, not only an inherited methylome.
The Hoxa9 sites MLL protects from methylation bind the CXXC domain 4- to 9-fold more tightly than matched unprotected sites.