Analysis (runnable): does the infant methylome predict HOXA status once age and breakpoint are controlled?
also called HOXA status predictor · methylome-breakpoint deconfounding
Using data that already exists, we can test whether a leukemia's HOXA gene activity is better explained by the chemical state of the cell it started in or by exactly where the chromosome broke. It is a computation, not an experiment.
The same thing, for a scientist
A multivariable analysis on existing infant and pediatric KMT2A-rearranged ALL cohorts to separate two confounded predictors of HOXA-cluster expression: KMT2A breakpoint intron and cell-of-origin epigenome. Inputs are DNA methylation arrays or WGBS, KMT2A breakpoint intron calls, HOXA-cluster expression and age at diagnosis. The model regresses HOXA status on methylation features with breakpoint and age as covariates, plus the reverse specification, and reports variance explained by each. Directly tests hyp-methylome-of-origin-sets-fusion-target-set and addresses the fact that the sole primary dataset supporting the HOXA split has age and breakpoint perfectly confounded.
What is being deconfounded, and with what data1 deeper
The only primary dataset reporting the HOXA split assigned infants to the HOXA-low/intron-11-12 group and non-infants to the HOXA-high/intron-9-10 group 1. Age and breakpoint therefore co-vary perfectly, and no causal claim is recoverable from it.
Three candidate data sources, all of which should already contain the needed variables:
- Interfant-99/06-derived resources. The 69-patient integrative methylome-transcriptome study of infant B-ALL used uniformly treated Interfant-99/06 patients and generated genome-wide methylation plus transcriptome, mapping methylome signatures onto normal haematopoietic chromatin states 2. HOXA expression is derivable from its transcriptome; breakpoint calls may need to be added.
- Nordlund pediatric ALL methylation cohort. 435,941 CpGs across 764 children at diagnosis with subtype annotation and outcome 3. Large, but KMT2A-r is a minority and the abstract reports no KMT2A-r-specific figures, so its value is as a methylation backbone and normalisation reference rather than as the primary KMT2A-r cohort.
- Stumpel/Rotterdam infant-ALL methylation series, which already reports partner-level methylation differences (t(4;11)/t(11;19) extensively hypermethylated; t(9;11) near-normal) 4 — the subtype structure the model must account for.
Model specification. Primary: HOXA-cluster score ~ methylation features (HOXA-locus CpGs; global/repetitive methylation; a methylation-derived cell-of-origin score) + breakpoint intron + age. Report partial R-squared per term with nested-model likelihood-ratio tests. Secondary (the reverse specification): HOXA score ~ breakpoint + age, then ask whether methylation features add explanatory power. Pre-register the feature set to avoid fishing across 450k probes; restrict to HOXA-locus CpGs, the published infant-ALL differentially methylated signature, and a small number of global-methylation summaries.
Interpretation rule, set in advance. If methylation features retain significant partial R-squared with breakpoint and age in the model, the methylome hypothesis survives. If breakpoint retains it and methylation does not, the breakpoint model wins and hyp-methylome-of-origin-sets-fusion-target-set should be marked refuted.
Honest assessment of feasibility
The binding constraint is not compute — it is whether any single cohort has all three of methylation, breakpoint intron and HOXA expression on the same patients. Methylation plus expression exists 2. Breakpoint intron calls exist in the Marschalek/DCAL and Interfant clinical datasets. Whether they are linkable at patient level in an accessible repository is the first thing to establish, and it is a one-week scoping task rather than a project.
If linkage proves impossible, the fallback is a within-infant analysis: among infants only (age controlled by restriction), does breakpoint intron still separate HOXA-high from HOXA-low? A null there would substantially weaken the breakpoint model at a fraction of the effort.
Status is planned. Nothing has been run; no result should be cited from this node.
| script | analyses/hoxa-methylome-deconfound/run.py |
|---|---|
| model | HOXA-cluster score ~ methylation features + breakpoint intron + age, with partial R-squared per term and nested likelihood-ratio tests; reverse specification run as a check; feature set pre-registered to HOXA-locus CpGs, the published infant-ALL DM signature and global-methylation summaries |
| fallback design | within-infant analysis restricting to infants only, asking whether breakpoint intron alone separates HOXA-high from HOXA-low |
| status | planned |
| result summary | Not yet run. No result may be cited from this node. |
| decision rule | methylation retains significant partial R-squared with breakpoint and age in the model -> methylome hypothesis survives; breakpoint retains it and methylation does not -> mark hyp-methylome-of-origin-sets-fusion-target-set refuted |
| feasibility blocker | whether methylation, breakpoint intron and HOXA expression are linkable at patient level in an accessible repository (one-week scoping task) |
| compute needs | trivial — laptop or a single Explorer CPU node |
| cost tier | low |
inputs
[ "Infant/pediatric KMT2A-r ALL DNA methylation arrays or WGBS (Interfant-99/06-derived; Nordlund pediatric ALL resource as backbone)", "KMT2A breakpoint intron calls per patient", "HOXA-cluster expression (HOXA5, A7, A9, A10) from matched RNA", "Age at diagnosis" ]
tests hypotheses
[ "hyp-methylome-of-origin-sets-fusion-target-set" ]
- Methylation measured in leukemic blasts reflects the leukemia, not the cell of origin, so a positive result is open to a reverse-causation reading: HOXA-high biology could be producing the methylation pattern rather than the reverse.
- With age and breakpoint strongly correlated in any real cohort, collinearity may leave all three terms individually non-significant — an ambiguous outcome that must be anticipated and reported as such rather than resolved by model selection.
- A refutation here would weaken the case that cell-of-origin epigenome drives fusion target choice, which several other nodes in this beat lean on.
- Are KMT2A breakpoint intron calls linkable at patient level to any published infant-ALL methylation cohort? Everything else in this analysis is straightforward; this is the single gate.
- Is n=69 adequate for a three-predictor model with correlated predictors? A power calculation should precede the analysis, and if it is not, the within-infant fallback becomes the primary design.
- Should the cell-of-origin score be derived from a fetal versus adult haematopoietic methylation reference rather than from the leukemia methylome itself? The latter risks circularity and the former requires a reference panel that may not exist for fetal progenitors.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In the sole primary dataset reporting the split, pediatric t(4;11) ALL divided 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 — leaving age and breakpoint fully confounded.
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, 2009 microarrays, signature derived against only three normal marrow samples. This is the dataset whose confound the analysis exists to break, not a source of the answer.
Evidence that argues against this (1)
- The confound is worse than age-versus-breakpoint: fusion partner is a third correlated variable and the HOXA/non-HOXA split is not binary. Multi-omics clustering of 84 infant KMT2A-rearranged leukaemias resolves FIVE integrative clusters whose identity tracks master transcription factor AND fusion partner together - IRX-type differentiated, IRX-type undifferentiated, HOXA-type MLLT1, HOXA-type MLLT3, HOXA-type AFF1. Any model regressing HOXA status on methylome with age and breakpoint as covariates must also carry partner, and the outcome variable should be cluster membership rather than a HOXA-high/low dichotomy, or the analysis will recover partner identity and call it methylome. Isobe T et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Genome-wide DNA methylome and transcriptome data exist for 69 patients with de novo KMT2A-rearranged and non-rearranged infant B-ALL uniformly treated on Interfant-99/06, with methylome signatures mapping to enhancer and transcriptional-control chromatin states of normal haematopoietic cells and segregating KMT2A-r from non-KMT2A-r.
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
The paper does not analyse HOXA status or KMT2A breakpoint position; whether breakpoint intron calls can be linked to these patients is the key unresolved feasibility question. n=69 across two molecular groups limits power for a three-predictor model.
Evidence that argues against this (1)
- Mapping methylome signatures onto enhancer and transcriptional-control chromatin states of NORMAL haematopoietic cells is the step that decides the interpretation, and it is where a cell-of-origin explanation and a fusion-driven explanation become indistinguishable. Infant blasts derive from a fetal progenitor whose methylome differs from the adult reference the states were called in, so differences attributed to the rearrangement may be differences in developmental origin. Oncofusion knowledge graph (internal analysis) 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
A pediatric ALL methylation resource covering 435,941 CpG sites in 764 children at diagnosis and 27 at relapse defines pan-ALL and subtype-specific signatures, including 9,406 shared predominantly hypermethylated CpGs and 6,612 CpGs predominantly hypermethylated at relapse, with subtype-specific CpGs stratifying relapse-free survival.
Genome-wide signatures of differential DNA methylation in pediatric acute lymphoblastic leukemia · Nordlund J, Backlin CL, Wahlberg P, et al. (Syvanen AC, senior author) · Genome Biology 14(9):r105 2013What this evidence cannot show
Pan-ALL numbers only; the abstract reports no KMT2A-r-specific CpG counts, so these figures must not be attributed to KMT2A-r. Predominantly non-infant cohort, so useful as a methylation backbone rather than as the KMT2A-r test set.
Evidence that argues against this (1)
- This is a pan-ALL resource and its KMT2A contribution is a subtype slice of 764 children, most of whom have other cytogenetics. The 27 relapse samples are the weak part: 6,612 CpGs called as relapse-hypermethylated from 27 paired-era samples, across 435,941 sites, is a multiple-testing problem, and relapse samples differ from diagnosis in blast purity and in prior treatment as well as in biology. Leukemia 2021
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Promoter methylation separates 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 resemble normal bone marrow, and the degree of hypermethylation within t(4;11)/t(11;19) influences relapse-free survival.
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 platform; the partner-methylation association is itself confounded with age, which the model must handle rather than inherit.
Evidence that argues against this (1)
- The entry's own content undercuts the fusion-centric reading: t(9;11)-positive infant ALL resembles NORMAL bone marrow while t(4;11) and t(11;19) are extensively hypermethylated. If the rearrangement drove the methylome, the three rearranged groups would not split with one of them landing on the normal side. Partner identity, or the cell the partner's fusion arises in, is doing the work. Oncofusion knowledge graph (internal analysis) 2026
blocked on the same agreement