In infant ALL, IRX1 transcript is enriched in KMT2A::AFF1 (21 of 48) and present but rare outside it (3 of 26) - so the occupancy zero outside AFF1 is not an expression zero
also called q149 GSE68720 IRX1 by partner · IRX1 expression AF4 vs non-AF4 infant ALL
In 97 infant leukaemia patients, the IRX1 gene is switched on in about 44% of those whose fusion partner is AFF1, but in only about 12% of those with a different partner. It is not completely off outside AFF1 - a few children with other partners have IRX1 as high as anyone.
The same thing, for a scientist
**In infant B-lineage ALL (Children's Oncology Group P9407, GSE68720, n=97 primary patients, Affymetrix GPL570), IRX1 transcript is called ON in 21 of 48 KMT2A::AFF1 cases (43.8%, Wilson 95% 30.7-57.7%) and in 3 of 26 non-AFF1 KMT2A-rearranged cases (11.5%, Wilson 95% 4.0-29.0%)** - Fisher exact **two-sided p = 0.0047** (one-sided greater 0.0039), Mann-Whitney on the continuous values two-sided p = 0.0049, permutation over 20,000 label shuffles two-sided p = 0.0018 (one-sided greater 0.0009) [E1]. **This is a different assay in different patients from the occupancy result it bears on.** `ana-q147-irx1-fusion-occupancy-is-confined-to-kmt2a-aff1` found the KMT2A-N oncoprotein at IRX1 in **0 of 18** non-AFF1 samples of GSE252378. This entry says the transcript is present in **3 of 26** non-AFF1 infant ALL cases of a *different* cohort. **No patient appears in both.** The pre-registered branch therefore fired as **NEITHER / UNINFORMATIVE for the binary question as posed**, and the informative statements are the ones below. **Three things that are true at once, and the brief's dichotomy cannot hold all three:** 1. IRX1 is **not silent** outside AFF1. Two of three KMT2A::EPS15 cases sit at 10.11 and 11.07 log2 units against a per-sample floor of 6.41 and 6.75, i.e. as high as the highest AFF1 cases (max 11.37). That is branch **(B) EXPRESSED-BUT-UNOCCUPIED** instantiated, as an existence claim across cohorts. 2. IRX1 **is** strongly AFF1-enriched, which is branch **(A) SILENT** for the majority of non-AFF1 cases. 3. The non-AFF1 KMT2A-r rate (3/26) is **statistically indistinguishable from the KMT2A-germline infant ALL controls in the same cohort** (3 of 17, 17.6%; Fisher two-sided p = 0.666). So outside AFF1, having a KMT2A rearrangement does not change IRX1 status at all relative to infant ALL generally. **Threshold sensitivity, because this project has been burned by it.** Under four floors - max, 95th percentile, median and min of the same ten-gene silent panel - the non-AFF1 rate is 11.5%, 11.5%, 15.4% and 26.9%, giving NEITHER, NEITHER, NEITHER and branch B. The AFF1-versus-non-AFF1 *contrast* survives all four; the absolute on/off verdict does not. Report the contrast, not the rate.
The table, with n stated and the small groups NAMED rather than computed
GSE68720, per-sample mll_par: field, IRX1 probe 230472_at, pre-registered floor = max of the ten-gene silent panel in that same sample.
mll_par: | current symbol | n | IRX1 ON | rate | Wilson 95% |
|---|---|---|---|---|---|
| AF4 | AFF1 | 48 | 21 | 43.8% | 30.7-57.7% |
| ENL | MLLT1 | 16 | 1 | 6.3% | reported, n>=5 |
| AF9 | MLLT3 | 6 | 0 | 0% | rule of three: <50% |
| EPS15 | EPS15 | 3 | 2 | NAMED, NOT COMPUTED - values 3.24, 10.11, 11.07 | |
| ASAH3 | unresolved retired symbol | 1 | 0 | NAMED, NOT COMPUTED - value 2.78 | |
| partner unknown | - | 6 | 1 | 16.7%, value 7.07 | |
| KMT2A germline | control, same cohort, same array | 17 | 3 | 17.6% - this is the NULL |
Pooled non-AFF1 KMT2A-r (AF4, germline and partner-unknown excluded): 3 of 26 = 11.5%.
The three non-AFF1 ON cases, named: GSM1679800 (EPS15, IRX1 11.07, floor 6.75, karyotype 46,XY,T(1;11;3)(P32;Q23;P21)), GSM1679782 (EPS15, 10.11, floor 6.41, 46,XY,T(1;11)(P32;Q23)), GSM1679748 (ENL, 6.23, floor 5.99, T(11;19)(Q23;P13.3), and marginal). Plus GSM1679787 in the partner-unknown group at 7.07.
Why this cohort and not the others - the resource survey in one line each
GSE68720 is the only public resource found that carries a per-sample KMT2A partner beside an expression matrix in B-lineage disease. TARGET ALL Phase II (n=1,978) carries MLL_STATUS: Positive/Negative and MOLECULAR_SUBTYPE: MLL-Rearranged and no partner; TARGET AML carries MLL: Yes/No; bll_target_gdc and aml_target_gdc carry 20 clinical attributes with no cytogenetic field at all; MILE stage 1 (GSE13159, n=2,096) says only leukemia class: Pro-B-ALL with t(11q23)/MLL (n=70) and AML with t(11q23)/MLL (n=38) - the band, never the partner. Those are verified-absent for partner annotation, not unchecked. See absence-q149-only-one-public-b-lineage-cohort-annotates-the-kmt2a-partner-beside-an-expression-matrix.
And the series was found only through the RETIRED symbol AF4. Its metadata never says AFF1 and it does not appear in a GEO search on KMT2A. infant acute lymphoblastic leukemia AND AF4 returns it; ... AND KMT2A does not.
IRX2 rides along inside AFF1 and nowhere else - and it is this cohort's own prognostic gene
Measured on the same samples with the same floor: IRX2 is ON in 9 of 48 AF4 cases and in 0 of 16 ENL, 0 of 6 AF9, 0 of 3 EPS15, 0 of 6 partner-unknown and 0 of 17 KMT2A-germline cases. All 9 IRX2-ON cases are also IRX1-ON (a strict subset: 9 double-positive, 12 IRX1-only, 27 double-negative, 0 IRX2-only).
The cohort's own publication independently made IRX2 prognostic in these exact patients 2: IRX2 is one of three genes predictive of event-free survival beyond age and MLL status, and inside the MLL-AFF1 cases elevated IRX2 sits in the 7-gene classifier associated with worse EFS (5-year EFS 20% vs 65%). That is the same 97 samples, so it is not independent validation - it is the observation that the IRX1/IRX2-high AFF1 subgroup and Kang's poor-outcome AFF1 subgroup are drawn from one dataset and may be the same patients. Nobody has cross-tabulated them; the outcome variable is not in the GEO record.
Note also that the two IRX1-high EPS15 cases are IRX2-negative, so whatever switches IRX1 on in EPS15 is not the AFF1 programme.
What the numbers are in, and what the floor is
Values are the GSE68720 series-matrix values on GPL570 (log2 scale; cohort ACTB median 13.47, PTPRC median 10.48, silent-panel medians 3.46-6.20). Probes collapsed to the maximum across a gene's probes. IRX1 has exactly one probe, 230472_at, annotated iroquois homeobox 1 in GPL570.annot.gz fetched 2026-09-24.
Twelve multi-gene probes were dropped and are named in the results file - eight ACTG1///ACTB, two HOXA10-HOXA9///MIR196B///HOXA9, one MYH7///MYH6, one HOXA10-AS///MIR196B. A consequence worth stating: GPL570 has no single-gene HOXA9 probe at all, so the HOXA control in this cohort is HOXA10 (213147_at, 213150_at) and the substitution is named, not silent.
All 97 samples passed the PTPRC gate (PTPRC above the silent floor), so nothing was excluded for control failure and there is no unreported filtering.
| cohort | GSE68720, COG trial P9407 infant ALL, Affymetrix HG-U133 Plus 2.0 (GPL570) |
|---|---|
| n patients | 97 |
| assay | microarray expression, series matrix, log2 |
| irx1 probe | 230472_at |
| hoxa control probe | HOXA10 213147_at / 213150_at - GPL570 has NO single-gene HOXA9 probe |
| partner field | !Sample_characteristics_ch1 = mll_par: |
| found via retired symbol only | AF4; invisible to a GEO search on KMT2A |
| threshold rule | IRX1 ON iff IRX1 > max of the ten-gene silent panel (ALB INS MYH7 CRYAA RHO SFTPC KRT1 MYOG GFAP CDX2) in the SAME sample |
| preregistration | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/PREREGISTRATION.md |
| script | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/gse68720.py |
| results | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/results149_gse68720.json |
| prereg branch fired | NEITHER / UNINFORMATIVE for the binary; the AFF1-vs-non-AFF1 contrast is the informative result |
| absence class | measured-not-analysed |
partner symbols are RETIRED
{
"AF4": "AFF1",
"ENL": "MLLT1",
"AF9": "MLLT3"
}measurements — the numbers behind the chart
[
{
"quantity": "proportion_IRX1_ON",
"subject": "KMT2A::AFF1 infant ALL (mll_par AF4)",
"value": 0.4375,
"units": null,
"n": 48,
"universe": "GSE68720 97 infant ALL, GPL570, floor = max of 10-gene silent panel per sample",
"null": 0.1765,
"note": "21 of 48; Wilson 95% 0.307-0.577; null is the 17 KMT2A-germline controls in the same cohort"
},
{
"quantity": "proportion_IRX1_ON",
"subject": "non-AFF1 KMT2A-r infant ALL (ENL+AF9+EPS15+ASAH3)",
"value": 0.1154,
"units": null,
"n": 26,
"universe": "GSE68720 97 infant ALL, GPL570, floor = max of 10-gene silent panel per sample",
"null": 0.1765,
"note": "3 of 26; Wilson 95% 0.040-0.290; indistinguishable from the germline null, Fisher two-sided p=0.666"
},
{
"quantity": "fisher_exact_p_two_sided",
"subject": "AFF1 vs non-AFF1 KMT2A-r, IRX1 ON counts 21/48 vs 3/26",
"value": 0.004733,
"units": null,
"n": 74,
"universe": "GSE68720 infant ALL, pre-registered max floor",
"null": null,
"note": "one-sided greater 0.003865; Mann-Whitney on continuous values two-sided 0.004893; permutation two-sided 0.0018 over 20000 shuffles seed 149"
},
{
"quantity": "proportion_IRX2_ON",
"subject": "KMT2A::AFF1 infant ALL",
"value": 0.1875,
"units": null,
"n": 48,
"universe": "GSE68720, same floor, IRX2 probes",
"null": 0,
"note": "9 of 48; IRX2 ON is 0 in every non-AFF1 group and 0 of 17 germline controls; all 9 are also IRX1-ON"
},
{
"quantity": "max_IRX1_log2",
"subject": "KMT2A::EPS15 infant ALL cases GSM1679800 / GSM1679782",
"value": 11.07,
"units": "log2 GPL570 series-matrix units",
"n": 3,
"universe": "GSE68720; per-sample floors 6.75 and 6.41",
"null": null,
"note": "second case 10.11; AFF1 maximum is 11.37, so these are AFF1-level. n=3 - NAMED, NOT COMPUTED"
}
]- If IRX1 status is used to select patients for any IRX1- or MYC-directed intervention, this analysis says the selection would enrol mostly KMT2A::AFF1 children and would MISS the 3 of 26 non-AFF1 and the 3 of 17 KMT2A-germline infants who are equally IRX1-high. A partner-based proxy for IRX1 status is wrong in both directions.
- The threshold decides the enrolment. Under the four floors tried here the non-AFF1 'IRX1-positive' population is 3, 3, 4 or 7 of 26 children. A trial eligibility criterion written as 'IRX1-positive' without a stated assay and cut-point is not implementable.
- IRX1 is reported INACTIVATED in 16.2% of Down-syndrome myeloid leukaemia with consequent MYC-pathway upregulation (PMID 38513239), so an IRX1-lowering agent would be pushing in the direction that disease's driver already goes.
- Are Kang's IRX2-high poor-outcome KMT2A::AFF1 infants the same children as the IRX1-high ones? The cross-tab needs the P9407 outcome variable, which is not in the GEO record, and it would connect the iALL-IRX1 subgroup to survival for the first time in this corpus.
- Why are 2 of 3 KMT2A::EPS15 infant ALL cases IRX1-high at AFF1 magnitude while IRX2-negative? If IRX1 without IRX2 is a different state from IRX1 with IRX2, the iALL-IRX1 subgroup is two things.
- One KMT2A-germline infant ALL case reaches IRX1 10.37. What is that leukaemia? If IRX1-high infant ALL exists without any KMT2A rearrangement, IRX1 is not downstream of the fusion at all in that patient.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In 97 primary infant ALL samples from COG P9407 profiled on Affymetrix GPL570 and deposited as GSE68720, IRX1 transcript exceeds a per-sample ten-gene silent-tissue floor in 21 of 48 KMT2A::AFF1 cases and in 3 of 26 non-AFF1 KMT2A-rearranged cases, Fisher exact two-sided p = 0.0047.
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
This is MY analysis of their deposited data, not their published result - Kang et al. never report IRX1 by partner. Array, not RNA-seq: a single probe, no isoform resolution, and an array floor is a hybridisation background rather than a zero. The 'ON' call depends on a threshold and the pre-registered threshold gives 11.5% while the most permissive of four gives 26.9%; only the AFF1-versus-non-AFF1 contrast is stable across all four. Partner comes from the submitter's `mll_par:` field, which I did not independently verify against the fusion transcript; the `karyotype:` field agrees where it is populated but is `--` in 29 of 97. Non-AFF1 n=26 is small and the three ON cases are two EPS15 and one marginal ENL.
Evidence that argues against this (1)
- The non-AFF1 KMT2A-r rate (3/26, 11.5%) is not distinguishable from the KMT2A-GERMLINE infant ALL controls in the same cohort on the same array (3/17, 17.6%; Fisher two-sided p = 0.666), and one germline case reaches 10.37. So IRX1-high is not a KMT2A-rearrangement phenotype outside AFF1 - it occurs in infant ALL without any KMT2A rearrangement at the same rate. A reader taking '3 of 26 non-AFF1 cases express IRX1' as evidence about KMT2A biology would be over-reading it. Kang H et al. 2012
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In the same 97-patient COG P9407 infant ALL cohort, IRX2 is one of three genes predictive of event-free survival beyond age and MLL status, and within MLL-AFF1 cases elevated IRX2 expression is part of a 7-gene classifier correlated with worse EFS (5-year EFS 20% versus 65%).
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
Read from the abstract, which states the classifier and the EFS split verbatim; the per-patient IRX2 values behind it were not re-derived from their supplementary tables. The link drawn here - that Kang's IRX2-high poor-outcome AFF1 patients and this beat's IRX1-high AFF1 patients may be the same children - is a hypothesis generated by noticing IRX2-ON is a strict subset of IRX1-ON in my own re-analysis; the outcome variable is not in the GEO record and the cross-tab has NOT been done.
Evidence that argues against this (1)
- IRX2 is a different gene from IRX1 and sits at a different position in the IRX cluster; a prognostic association for IRX2 is not an association for IRX1, and this project has already recorded that completing a logical pair is fabrication. The nesting I measured (9 of 9 IRX2-ON are IRX1-ON) is real, but 12 of the 21 IRX1-ON AFF1 cases are IRX2-negative and would be invisible to Kang's classifier. Kang H et al. 2012
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
The published IRX1 mechanism in this disease is framed as t(4;11)-specific, and the partner-restricted expression enrichment measured here is consistent with that framing while extending it from a mechanism proposal to a 74-patient partner-stratified count.
The IRX1/HOXA connection: insights into a novel t(4;11)- specific cancer mechanism · Kuhn A, Loscher D, Marschalek R · Oncotarget 7:35341-35352 2016What this evidence cannot show
Kuhn et al. never tested the other partners - the t(4;11)-specific framing in their title is an assertion about the system they worked in, not a measured comparison. This entry is the first partner-stratified expression count in this corpus, and it supports their framing by a route they did not take; it does not confirm their mechanism.
Evidence that argues against this (1)
- Two of three KMT2A::EPS15 cases here are IRX1-high at AFF1-level magnitude, which is a direct counter-example to a strictly t(4;11)-specific reading of IRX1 activation. n=3 and named, not computed - but it is the observation that most threatens the specificity claim. Kang H et al. 2012
- supports →IRX1 fusion occupancy is confined to KMT2A::AFF1: 0 of 18 samples across five other partners
expression enrichment in AFF1 by a different assay in a different cohort; but it also shows the occupancy zero is not an expression zero
- related to →IRX1 — Iroquois homeobox 1, the gene that names the HOXA-negative half of infant t(4;11) ALL
first partner-stratified IRX1 transcript count in this corpus, 74 partner-named infant ALL cases
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
IRX1 ON in 21 of 48 KMT2A::AFF1 infant ALL cases, 43.8%
the two assays are in different patients and give different answers; this is the worked case
COG P9407, 97 infant ALL patients
closes that entry's first open question in the B-lineage disease rather than in AML
the iALL-IRX1 subgroup is 43.8% of KMT2A::AFF1 infant ALL by this count
2 of 3 KMT2A::EPS15 infant ALL cases are IRX1-high at AFF1 magnitude - the counter-example to a t(4;11)-specific reading
the pre-registered binary verdict for this cohort is threshold-dependent; the contrast is not
the reason the B-lineage arm of this beat is one 97-patient 2012 array deposit
the cohort this analysis is computed on
the named n=3 group inside the partner table
the non-AFF1 rate matching the KMT2A-germline control is the second prediction
the IRX2 arm of the same partner-stratified measurement
the within-cohort KMT2A-germline null for that partner table
reproduces all six of its IRX1 quantities exactly, which is what makes the IRX2 disagreement interpretable