The HOXA/IRX1 mutual exclusivity is reproduced at the transcript level in 48 KMT2A::AFF1 infant ALL cases (Fisher two-sided p < 1e-6) and is ABSENT in KMT2A::MLLT1 and in KMT2A-germline controls
also called q149 iALL-HOXA9 vs iALL-IRX1 replication · HOXA10/IRX1 dichotomy by partner
Infant leukaemias with the AFF1 partner come in two flavours: HOXA-on/IRX1-off or IRX1-on/HOXA-off, almost never both. We reproduced that split in 48 patients. In leukaemias with a different partner, the split does not exist - they are simply HOXA-on.
The same thing, for a scientist
Symeonidou & Ottersbach described *"a remarkable mutually exclusive expression of the HOXA9/HOXA10 and IRX1 genes"* defining iALL-HOXA9 and iALL-IRX1 from two previously published RNA-seq datasets [E1]. **That dichotomy reproduces here, independently, from GSE68720 array data with a pre-registered threshold, and it is partner-restricted** [E2]: | group | n | HOXA+ IRX1- | HOXA- IRX1+ | both+ | both- | Fisher two-sided | |---|---:|---:|---:|---:|---:|---| | **KMT2A::AFF1 (`mll_par: AF4`)** | 48 | 23 | **19** | 2 | 4 | **< 1e-6** | | KMT2A::MLLT1 (`ENL`) | 16 | 12 | **0** | 1 | 3 | **1.0** | | KMT2A-germline infant ALL | 17 | 6 | 2 | 1 | 8 | **1.0** | **42 of 48 AFF1 cases are exactly one of the two states.** In MLLT1 there is no IRX1+ HOXA- case at all - the group is simply HOXA-high (13 of 16 HOXA-ON), with one double-positive. **This is a stronger and more direct version of the q147 occupancy finding.** q147 measured Spearman(HOXA, IRX1) = -0.714 on **8** KMT2A::AFF1 occupancy samples with a **two-sided** exact p of 0.0576 (the one-sided 0.0288 is the value that must not travel alone). Here the same anti-relation is a 2x2 on **48 patients** with a two-sided Fisher p below 1e-6, and the negative in MLLT1 is a clean n=16 rather than the n=6 of the occupancy arm. **And the dichotomy does NOT generalise to paediatric AML.** Over 171 non-AFF1 KMT2A-r paediatric AML samples in `aml_stjude_2024`, Spearman(HOXA9, IRX1) with average ranks is **+0.005, two-sided permutation p = 0.937** over 20,000 shuffles - no relation of either sign. In the 20 IRX1-high non-AFF1 AML samples HOXA9 is 6.8-9.4, i.e. high. **HOXA and IRX1 are co-expressed in myeloid KMT2A-r disease and mutually exclusive in AFF1 B-lineage disease.**
The control gene is HOXA10, and that is not a free choice
GPL570 has no single-gene HOXA9 probe. Both HOXA9 probes are annotated HOXA10-HOXA9///MIR196B///HOXA9 (209905_at, 214651_s_at) and were dropped by the multi-gene rule that also dropped eight ACTG1///ACTB probes. The HOXA readout in this cohort is therefore HOXA10 (213147_at, 213150_at), which is single-gene.
This is a substitution, not an equivalence - but it is the substitution Symeonidou & Ottersbach's own phrasing licenses, since their subgroup is defined on HOXA9/HOXA10 together, not HOXA9 alone. It is named here rather than left in a script.
The AML arm uses HOXA9 proper (entrez 3205) from the aml_stjude_2024_mrna_seq_cpm profile, so the two arms are not using the same gene. That is a reason the AML null and the ALL positive cannot be pooled, and they are not.
Why a 2x2 and not a correlation
A Spearman over IRX1 is tie-degenerate: most samples sit in a tied block at the detection floor, which is why q147 reported the attainable tie-break range and why concept-q147-a-spearman-over-exact-zeros-is-tie-degenerate-and-argsort-picks-one-value exists. The AML Spearman here was computed with average ranks and its tie structure is printed alongside it (IRX1: 11 tie groups, largest 3; HOXA9: no ties) - so it is interpretable, and it is flatly null.
For the ALL arm a 2x2 on the thresholded calls is the better instrument, because the published claim is about two subgroups, not about a monotone relation. Both tails of the Fisher are reported in the results file.
What would make this wrong
- The threshold. The 2x2 uses the pre-registered max-of-silent-panel floor for both genes. A different floor moves cells between rows. The AFF1 result is extreme enough (2 double-positives in 48) that it survives; the MLLT1 negative is the one to distrust, because with 16 cases a single reassignment changes the picture.
- HOXA10 is not HOXA9. If the AFF1 dichotomy is specifically HOXA9-versus-IRX1 and HOXA10 tracks it imperfectly, the 23/19 split is approximate.
- A shared platform floor is not a shared biology. Both genes being scored against the same per-sample floor induces a weak negative dependency by construction when a sample's floor is high. The AFF1-versus-MLLT1 contrast controls for this, because both groups are scored identically and only one shows the pattern.
| hoxa gene used in ALL arm | HOXA10 (213147_at, 213150_at) - GPL570 has NO single-gene HOXA9 probe |
|---|---|
| aml spearman hoxa9 irx1 | 0 |
| script | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/anchors149.py |
| results | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/results149_anchors.json |
| absence class | measured-not-analysed |
aff1 2x2
{
"HOXA_only": 23,
"IRX1_only": 19,
"both": 2,
"neither": 4,
"n": 48,
"fisher_two_sided": "<1e-6"
}mllt1 2x2 — the numbers behind the chart
{
"HOXA_only": 12,
"IRX1_only": 0,
"both": 1,
"neither": 3,
"n": 16,
"fisher_two_sided": 1
}germline 2x2 — the numbers behind the chart
{
"HOXA_only": 6,
"IRX1_only": 2,
"both": 1,
"neither": 8,
"n": 17,
"fisher_two_sided": 1
}measurements — the numbers behind the chart
[
{
"quantity": "fisher_exact_p_two_sided",
"subject": "HOXA10-ON x IRX1-ON 2x2 within KMT2A::AFF1 infant ALL",
"value": 0.000001,
"units": null,
"n": 48,
"universe": "GSE68720 GPL570, both genes scored against the same per-sample max-of-10-silent-genes floor",
"null": 1,
"note": "reported as <1e-6; the null cited is the same 2x2 in the 16 KMT2A::MLLT1 cases, Fisher two-sided 1.0"
},
{
"quantity": "count_IRX1only_HOXAonly",
"subject": "KMT2A::AFF1 infant ALL exclusive states",
"value": 42,
"units": "patients",
"n": 48,
"universe": "GSE68720; 19 IRX1-only + 23 HOXA-only",
"null": null,
"note": "2 double-positive, 4 double-negative"
},
{
"quantity": "spearman_rho_average_ranks",
"subject": "HOXA9 vs IRX1 transcript, non-AFF1 KMT2A-r paediatric AML",
"value": 0.005,
"units": null,
"n": 171,
"universe": "aml_stjude_2024 mrna_seq_cpm, all known-partner non-AFF1 KMT2A-r samples",
"null": 0,
"note": "two-sided permutation p 0.93705 over 20000 shuffles seed 149; one-sided greater 0.46853, less 0.53157; IRX1 has 11 tie groups, largest 3"
}
]- If iALL-HOXA9 and iALL-IRX1 respond differently to menin inhibition - which nothing yet measures - then a menin inhibitor trial that does not stratify on IRX1 is averaging over two diseases, and the 44% IRX1-high fraction measured here is large enough to move a response rate either way.
- The exclusivity is a reason to doubt any single-target strategy in this disease: a drug aimed at the HOXA arm has no purchase on 19 of 48 AFF1 infants, and one aimed at IRX1 has none on 23 of 48.
- Does the exclusivity hold with HOXA9 itself rather than HOXA10? An RNA-seq infant ALL cohort with partner annotation would settle it; GPL570 structurally cannot.
- Is there a single KMT2A::MLLT1 infant ALL case anywhere that is IRX1-high and HOXA-low? 0 of 16 here and 0 of 6 in the q147 occupancy arm; the two together are 0 of 22 across two assays, which is a rule-of-three bound below 13.6%.
- Within paediatric AML, does the HOXA9/IRX1 relation reappear inside a single FAB stratum? The pooled null over six morphologies could be hiding it, and this beat has already found one composition artefact in the same cohort.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Re-analysis of two previously published RNA-sequencing datasets identified a remarkable mutually exclusive expression of the HOXA9/HOXA10 and IRX1 genes in infant t(4;11) acute lymphoblastic leukaemia, defining two subgroups termed iALL-HOXA9 and iALL-IRX1.
HOXA9/IRX1 expression pattern defines two subgroups of infant MLL-AF4-driven acute lymphoblastic leukemia · Symeonidou V, Ottersbach K · Exp Hematol 93:38-43.e5 2021What this evidence cannot show
Read from the abstract, which states the mutual exclusivity and the subgroup names verbatim. Their analysis is of t(4;11) cases only - they did not test any other partner, so the partner-restriction reported here is not their claim and not their negative. Their n and their expression cut-point were not extracted.
Evidence that argues against this (1)
- Nagel's haematopoietic TALE-code places IRX1 in NORMAL pro-B cells, which would make iALL-IRX1 a cell-of-origin signature rather than an aberrant activation, and would make 'mutually exclusive subgroups' a statement about developmental stage rather than about the fusion. That disagreement is open and is recorded as dispute-is-irx1-expressed-in-normal-b-cell-development; it is NOT resolved here. Nagel S et al. 2022
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In 48 KMT2A::AFF1 infant ALL cases from GSE68720, IRX1 and HOXA10 transcript calls against a per-sample silent-tissue floor are mutually exclusive (23 HOXA-only, 19 IRX1-only, 2 both, 4 neither; Fisher exact two-sided p below 1e-6), while in 16 KMT2A::MLLT1 cases from the same cohort there is no IRX1-only case at all (Fisher two-sided p = 1.0).
Gene expression profiles predictive of outcome and age in infant acute lymphoblastic leukemia: a Children's Oncology Group study · Kang H, Wilson CS, Harvey RC, et al. · Blood 119:1872-1881 2012What this evidence cannot show
My analysis of their deposited data, not their published result. HOXA10 substitutes for HOXA9 because GPL570 has no single-gene HOXA9 probe. The MLLT1 negative rests on 16 cases and 3 double-negatives; it is a real absence of the IRX1-only state in that group, not a demonstration that the state cannot occur. Array, single probe per gene for IRX1, thresholded calls.
Evidence that argues against this (1)
- Scoring two genes against the same per-sample floor can manufacture a weak apparent exclusivity, because a sample with a high floor tends to be called negative for both - which populates the both-negative cell, not the two exclusive cells. In the AFF1 group the both-negative cell holds only 4 of 48, so the mechanism does not account for the result; but a reader should know the artefact exists and check the both-negative count before trusting any such 2x2. Kang H et al. 2012
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Over 171 non-AFF1 KMT2A-rearranged paediatric AML samples, Spearman's rho between HOXA9 and IRX1 transcript with average ranks is +0.005 with a two-sided permutation p of 0.937, so the HOXA/IRX1 anti-relation does not exist in myeloid KMT2A-rearranged disease.
A new genomic framework to categorize pediatric acute myeloid leukemia · Umeda M, Ma J, Westover T, et al. · Nat Genet 56:281-293 2024What this evidence cannot show
A null over a variable with a large tied block at the detection floor; the tie structure is printed with the rho (IRX1: 11 tie groups, largest 3). Different disease (AML), different platform, different gene for the HOXA arm (HOXA9 proper here, HOXA10 in the ALL arm) - so the ALL positive and the AML null are not a matched comparison and were not pooled. The absence of an anti-relation in AML is compatible with the AFF1-specific dichotomy being real, because there is no AFF1 arm in this AML cohort worth the name (n=1).
Evidence that argues against this (1)
- In paediatric AML the IRX1-high samples are overwhelmingly FAB M7, which is a lineage subgroup with its own HOXA behaviour; a null pooled over six morphologies can hide opposite relations inside them. The rho was NOT recomputed within FAB strata, so the AML null is a pooled null and could be a composition artefact in exactly the way this beat's own FAB analysis shows the on/off rate was. Umeda M et al. 2024
the same partner-restriction at the transcript level in 48 patients instead of 8 occupancy samples
- supports →iALL-HOXA9 — the HOXA-high state of infant KMT2A::AFF1 ALL, and the one everything is modelled on
23 of 48 KMT2A::AFF1 infant ALL cases are HOXA-only, independently re-derived
19 of 48 KMT2A::AFF1 infant ALL cases are IRX1-only, independently re-derived
0 of 16 KMT2A::MLLT1 infant ALL cases are IRX1-only, extending the occupancy negative to expression
- related to →Is IRX1 expressed in normal blood at all? Two of the three papers say no, one says exclusively yes
bears on the dispute without resolving it: the dichotomy is real but says nothing about normal pro-B cells
HOXA10 is the readout because GPL570 has no single-gene HOXA9 probe
the 20 IRX1-high AML samples are all HOXA9-high, which is why the dichotomy is absent here
the 2x2 on thresholded calls is used precisely because the rate is threshold-dependent
the cohort this 2x2 is computed on
PROM1 inherits its anti-relation to IRX1 from the HOXA/IRX1 exclusivity