IRX1 fusion occupancy is confined to KMT2A::AFF1: 0 of 18 samples across five other partners
In leukaemias driven by a broken KMT2A gene, the fusion protein sits on the HOXA genes whichever partner it is joined to. It sits on IRX1 only when the partner is AFF1 - in 18 samples carrying five other partners it never does.
The same thing, for a scientist
🔴 **LINEAGE CAVEAT added cycle 146 (`ana-q158-partner-exclusivity-does-discriminate-and-the-one-rival-that-fired-is-a-lineage-artefact`).** The gene-body AFF1-exclusive class survives a lineage-matched test - 7 against a null of 1.66, p 0.0044, using only the 7 B-lineage AFF1 samples against the 16-sample B-lineage pool - **but IRX1 is not one of its members.** IRX1's three positive samples are `p279`, `p318` and `p454`, and **`p318` is AML**, so within B-lineage IRX1 has 2 of 7 and falls below the k>=3 recurrence bar. **On the most lineage-controlled version of this test IRX1 cannot appear.** Its exclusivity (0 of 18 non-AFF1) is unaffected and holds at every gate tested; what is not lineage-robust is its *recurrence*. Independently, IRX1's own permutation p is 0.0219 unstratified but 0.0772 (patient), 0.1190 (institution, exact) and 0.1756 (B-lineage). 🟠 **SCOPED 2026-09-24 (cycle 135) by q149: the 0-of-18 cannot cover the subgroup where IRX1 is actually expressed.** q149 went to partner-annotated expression cohorts and found that **in AML, IRX1 expression is a FAB-M7 megakaryoblastic property and not a fusion property**: 13 of 20 KMT2A-r M7 cases (65.0%) against 2 of 96 M5 (2.1%) - but **54 of 97 non-KMT2A-r M7 cases (55.7%) are also positive**, and KMT2A-r M7 vs non-KMT2A-r M7 is **Fisher two-sided p = 0.47**. The fusion adds nothing; the morphology is the variable. 🔴 **And GSE252378 contains no megakaryoblastic sample at all.** I grepped its metadata: **zero** matches for `megakary`, `AMKL` or `erythro` across all 171 GSMs, whose source strings are pediatric AML, infant B-ALL, infant Pre-B-ALL, infant T-ALL and two lineage-switching cases. **So this entry's 0-of-18 is a bound over the partners present in one deposit, and that deposit systematically excludes the one morphology in which IRX1 is switched on.** 🔵 **What still stands:** within the samples actually profiled, the zero is a measurement and not a failure to measure - **17 of those 18 are HOXA-occupied at 2.94-29.11x in the same libraries.** The claim should be read as *IRX1 fusion-occupancy is confined to KMT2A::AFF1 among the lymphoid and monocytic blasts in GSE252378*, not as a statement about KMT2A-r leukaemia in general. 🔵 **And the expression side does not simply mirror it.** In infant ALL (GSE68720, 97 patients with per-sample partner calls) IRX1 is ON in **21 of 48 AFF1** against **3 of 26 non-AFF1** (Fisher two-sided **p = 0.0047**, which I reproduced exactly) - but the non-AFF1 rate is **indistinguishable from the 17 KMT2A-germline controls in the same cohort** (3 of 17, p = 0.67), and **AFF1 vs germline does not itself clear 0.05** (p = 0.0798). So expression is partner-enriched and the proper baseline arm is underpowered. --- Pre-registered extension of q145 from one fusion partner to all six in GSE252378 (27 samples, 24 independent patients, 133 KMT2A-N/C and IgG libraries, one protocol, one genome build, four institutions). **IRX1 is fusion-OCCUPIED in 3 of 8 KMT2A::AFF1 samples (p279 6.08x, p454 6.35x, p318 2.68x at gene-body scale) and in 0 of 18 samples carrying KMT2A-ENL (n=6), -AFDN (4), -AF10 (3), -AF9 (3) or -EPS15 (2)** [E1]. The highest IRX1 gene-body share anywhere outside AFF1 is **0.269x** an average promoter (A95941, KMT2A-AF9) - roughly a quarter of an average promoter and at or below that library's own cannot-be-occupied floor [E1]. **The libraries are demonstrably able to report occupancy: HOXA9/HOXA10 is OCCUPIED in 17 of the same 18 samples, at 5.7-29.1x** [E2]. The pre-registered AFF1-SPECIFIC branch fires on all three of its conditions and the GENERAL branch fails on its reciprocal-phenotype condition: **no non-AFF1 sample is IRX1-positive and HOXA-negative** [E1]. Rule of three: the true rate of IRX1 occupancy in non-AFF1 KMT2A-r is **below 16.7% with 95% confidence** [E1]. The published mechanism paper reached the same restriction by a different route eight years earlier - its title calls the IRX1/HOXA connection *a novel **t(4;11)-specific** cancer mechanism* [E3].
What was measured, and at which scale
Two scales, because q145's headline is at gene-body scale and a negative measured only at a narrower window would be a result about the ruler.
| scale | HOXA interval | IRX1 interval |
|---|---|---|
| promoter | HOXA9 and HOXA10 TSS +/- 1 kb (2,000 bp each) | IRX1 TSS +/- 1 kb (2,000 bp) |
| gene body | HOXA9-HOXA10 span, 17,883 bp | IRX1 gene, 5,572 bp |
Both give the same answer. At promoter scale IRX1 is OCCUPIED in p279 and p454 only; at gene scale p318 joins them. No non-AFF1 sample is IRX1-occupied at either scale.
Every interval was asserted for width, strand-correct TSS and containment before any value was read, because q145's own gate caught a programmatically derived HOXA10.txStart .. HOXA9.txEnd window - both genes minus-strand - that is 933 bp and overlaps neither gene, with no error raised. The gate here re-asserts 19,404 promoters at exactly 2,000 bp and re-checks 25 named loci against the RefSeq record.
Per partner, with the groups that cannot support a correlation named as such
| partner | untreated n | depth-adequate | HOXA occupied | IRX1 occupied | max IRX1 share |
|---|---|---|---|---|---|
| KMT2A-AFF1 | 8 | 5 | 4/8 | 3/8 | 6.350x |
| KMT2A-ENL | 6 | 5 | 6/6 | 0/6 | 0.246x |
| KMT2A-AFDN | 4 | 3 | 4/4 | 0/4 | 0.091x |
| KMT2A-AF10 | 3 | 2 | 2/3 | 0/3 | 0.249x |
| KMT2A-AF9 | 3 | 1 | 3/3 | 0/3 | 0.269x |
| KMT2A-EPS15 | 2 | 2 | 2/2 | 0/2 | 0.109x |
Four of the six groups cannot support a correlation at any effect size and none was computed for them: at n = 4 the smallest attainable exact two-sided permutation p is 2/4! = 0.083, at n = 3 it is 0.33, at n = 2 it is 1.0. Only AFF1 (n = 8) and ENL (n = 6) were correlated, and ENL's rho is +0.200, two-sided exact p = 0.714 - the direction does not even point the same way. See ana-q147-the-hoxa-irx1-anticorrelation-does-not-reproduce-in-kmt2a-enl.
The AF9 row is the weakest: only one of its three samples is depth-adequate.
Why this is not a depth artefact
Depth was gated before the axis was read. p_dz, the chance a promoter reads zero in both KMT2A-N antibodies, ranges 0.000-0.493 across the 27 samples; the pre-declared cuts were ADEQUATE <= 0.10, MARGINAL <= 0.20, INADEQUATE above. A zero in a shallow library is recorded as UNMEASURED, a third state, never as an absence.
Twelve untreated non-AFF1 samples are depth-adequate and pass the positive-control gate, against a pre-declared uninformative threshold of six. Across the pooled non-AFF1 samples Spearman(p_dz, HOXA) = +0.489 and Spearman(p_dz, IRX1) = -0.549, both below the pre-declared |rho| >= 0.6 that would have meant the ordering was depth rather than biology.
And the strongest single counter to a depth reading is that HOXA is occupied at 5.7-29.1x in the very same libraries in which IRX1 reads 0.00-0.27x.
What this does NOT show
- Not an expression claim. The published subgroup (Symeonidou & Ottersbach) is Kallisto -> tximport -> DESeq2 over two RNA-seq cohorts with no chromatin experiment of any kind. GSE252378 contains no RNA-seq. IRX1 expression is separately reported in AML subsets and in AML cell lines by Nagel and colleagues, so 'no fusion at IRX1' is not 'no IRX1' 4. See
ana-q147-irx1-expression-is-reported-in-aml-while-irx1-occupancy-is-not. - Not partner-causal. All three IRX1-occupied samples are KMT2A::AFF1 and from one institution and lineage-switch-associated. Partner, institution and cell state are confounded in this deposit and cannot be separated inside it. The competing reading is written out as
hyp-q147-irx1-occupancy-is-a-lineage-switch-cell-state-confounded-with-one-institution. - Not a claim about IRX1 protein. This is where the KMT2A N-terminal fragment sits, nothing more.
| deposit | GSE252378 |
|---|---|
| n samples | 27 |
| n independent patients | 24 |
| n libraries scored | 133 |
| n promoters | 19404 |
| genome build | hg38 |
| assay | AutoCUT&RUN, KMT2A-N1/N2 (fusion side), KMT2A-C1/C2, IgG |
| units | within-sample share (1.0 = an average promoter in that library) |
| floor rule | >=2.0x in both KMT2A-N antibodies AND above the 95th percentile of eight width-matched cannot-be-occupied loci |
| nonaff1 irx1 occupied | 0 |
| nonaff1 n | 18 |
| nonaff1 rule of three upper PCT | 16.7 |
| max nonaff1 irx1 gene share | 0.269 |
| hoxa occupied nonaff1 | 17/18 |
| hoxa share range nonaff1 | 0.87x - 29.11x (ALL 18 non-AFF1 samples) |
| spearman hoxa vs irx1 AFF1 gene scale | -0.7143 |
| exact p two sided AFF1 | 0.0576 |
| exact p one sided AFF1 | 0.0288 |
| spearman hoxa vs irx1 ENL gene scale | 0.2 |
| exact p two sided ENL | 0.7139 |
| preregistration | analyses/q147-hoxa-irx1-across-partners/PREREGISTRATION.md |
| prereg branch fired | AFF1-SPECIFIC |
| hoxa share range occupied nonaff1 | 2.94x - 29.11x (the 17 that clear the floor) |
| corrected cycle 135 | the single range key was being requoted as the OCCUPIED range; 0.87x is A73051, unoccupied. Both ranges are now named explicitly. |
| scope limit cycle 135 | GSE252378 contains no megakaryoblastic (FAB-M7) sample - zero metadata matches for megakary/AMKL/erythro across 171 GSMs - and q149 shows IRX1 expression in AML is an M7 property independent of the fusion (KMT2A-r M7 65.0% vs non-KMT2A-r M7 55.7%, p=0.47). The 0-of-18 therefore cannot cover the subgroup where IRX1 is on. |
irx1 occupied samples
[ "p279", "p318", "p454" ]
irx1 occupied partners
[ "KMT2A-AFF1" ]
analyses
[ "ana-q145-irx1-is-fusion-occupied-exactly-where-hoxa-is-not" ]
lineage matched class membership
{
"in_lineage_matched_class": false,
"reason": "p318 is AML; IRX1 has 2 of 7 B-lineage AFF1 samples, below k>=3",
"lineage_matched_members": [
"ELF1",
"GNAQ",
"H3C15",
"HSPA2",
"MIDN",
"PAN3",
"SPAG7"
],
"class_survives_lineage_matching": "7 vs null 1.66, p_upper 0.0044",
"irx1_exclusivity_unaffected": "0 of 18 non-AFF1, stable at every gate 1.25-3.0",
"by_entry": "ana-q158-partner-exclusivity-does-discriminate-and-the-one-rival-that-fired-is-a-lineage-artefact"
}- If IRX1 occupancy is AFF1-restricted, a menin inhibitor cannot be expected to act on it the way it acts on HOXA in every partner: the shared HOXA address is the menin-interface programme and the IRX1 address is not obviously part of it. A trial stratified on HOXA response could miss an IRX1-driven resistant compartment entirely.
- The one sample in this deposit sampled during menin inhibition (152985) loses HOXA occupancy AND fails the positive-control gate, i.e. the oncoprotein is globally off chromatin. IRX1 does not rise there - but that library cannot report anything, so 'IRX1 is not a menin-inhibitor escape route' is NOT established and must not be inferred from it.
- Targeting IRX1 itself is hazardous in both directions: IRX1 is a normal pro-B and megakaryocyte-erythroid progenitor factor, and its LOSS is a recurrent driver event in Down-syndrome myeloid leukaemia. An IRX1-degrading strategy would be acting on a gene whose inactivation is itself leukaemogenic in a related paediatric disease.
- Is IRX1 TRANSCRIBED in any of the 18 non-AFF1 samples? GSE252378 has no RNA-seq, so a sample could be IRX1-high and oncoprotein-free at IRX1, which would separate the expression subgroup from the occupancy subgroup completely.
- Does an untreated KMT2A-ENL lineage-switching blast - the design GSE252378 contains only under a menin inhibitor - acquire IRX1 occupancy? That single sample would decide between the partner reading and the cell-state reading.
- Which part of AFF1 would be required? The KMT2A N-terminal fragment, including the menin and LEDGF interfaces and the CXXC domain, is identical in all six fusions, so an AFF1-restricted address must come from the C-terminal partner or from something it recruits.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In GSE252378, KMT2A-N oncoprotein occupancy at IRX1 clears the pre-registered absolute floor (>=2.0x an average promoter in both N antibodies and above a width-matched cannot-be-occupied floor) in 3 of 8 KMT2A::AFF1 samples and in 0 of 18 untreated samples carrying KMT2A-ENL, -AFDN, -AF10, -AF9 or -EPS15; the largest non-AFF1 IRX1 gene-body share is 0.269x. Rule of three: true rate below 16.7% with 95% confidence.
q147 re-analysis of GSE252378: HOXA and IRX1 oncoprotein occupancy across all six KMT2A fusion partners, 27 samples, 133 libraries (analyses/q147-hoxa-irx1-across-partners) · Iqbal Lab oncofusion knowledge graph, beat q147 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
n = 18 non-AFF1 samples, 16 independent patients, one deposit, one assay, one genome build. Six of the 18 are depth-MARGINAL or INADEQUATE and four of those have IRX1 UNMEASURED (zero in both N antibodies at inadequate depth) rather than measured-absent; the 0/18 therefore rests on 12 fully adequate samples plus 6 weaker ones, and the rule-of-three bound computed on the 12 alone would be 25.0% rather than 16.7%. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean coverage over all 19,404 hg38 RefSeq promoters, so 1.0 = an average promoter in that library. GSE252378 has no spike-in, so no between-library absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels. The occupancy call is absolute, not rank-based: >=2.0x an average promoter in BOTH KMT2A-N antibodies AND above that library's own floor, where the floor is the 95th percentile of eight width-matched windows at loci that cannot be occupied in a leukaemia (ALB, RHO, MYH7, MYOD1, NEUROD1, KRT14, GFAP, INS). IgG is not the gate: q145 measured GSE252378 IgG libraries as 57-94% zero at promoters, and one of them (TB051939) reads 13.1x an average promoter at a locus that cannot be occupied.
Evidence that argues against this (2)
- IRX1 expression - not occupancy - is reported as aberrantly active in subsets of AML and in megakaryoblastic and myelomonocytic AML cell lines, so a gene that is on can be absent from this measurement. The two quantities are not interchangeable and this entry measures only the second. Nagel S et al. 2022
- The deposit's own paper reports that oncoprotein genomic enrichment is highly variable between samples and dynamically regulated, and that binding patterns over lineage gene sets are fusion-partner-specific - so a per-sample occupancy difference can reflect oncoprotein level or cell state rather than partner identity. Janssens DH et al. 2024
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
HOXA9/HOXA10 is fusion-OCCUPIED in 17 of the same 18 untreated non-AFF1 samples, at 5.7-29.1x an average promoter at gene-body scale, demonstrating that these libraries report oncoprotein occupancy at a comparable homeobox locus in the same nuclei in which IRX1 reads 0.00-0.27x.
q147 re-analysis of GSE252378: HOXA and IRX1 oncoprotein occupancy across all six KMT2A fusion partners, 27 samples, 133 libraries (analyses/q147-hoxa-irx1-across-partners) · Iqbal Lab oncofusion knowledge graph, beat q147 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
The one exception, A73051 (KMT2A-AF10), is HOXA-negative AND IRX1-negative at adequate depth and is treated as a finding rather than a failure - see `ana-q147-a73051-is-a-hoxa-negative-irx1-negative-kmt2a-af10-blast`. A73051 also has the weakest positive controls in the set (MEIS1 2.4-2.6x, CDK6 0.6-0.7x), so it passes the readability gate on MEIS1 alone. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean coverage over all 19,404 hg38 RefSeq promoters, so 1.0 = an average promoter in that library. GSE252378 has no spike-in, so no between-library absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- HOXA occupancy being present everywhere is consistent with, but does not establish, a menin/CXXC-driven shared address: the deposit paper shows the same oncoprotein-target gene sets are activated to very different degrees depending on oncoprotein level, so 'occupied' at 5.7x and at 29.1x are not the same biological state. Janssens DH et al. 2024
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
The published mechanistic study of the IRX1/HOXA relationship frames it as t(4;11)-specific in its own title, and reports IRX1 binding the MLL-AF4 complex at target-gene promoters and counteracting its promoter-activating function, with induction of HOXB4 and EGR family members.
The IRX1/HOXA connection: insights into a novel t(4;11)-specific cancer mechanism · Kuhn A, Loscher D, Marschalek R · Oncotarget 7:35341-52 2016What this evidence cannot show
Kuhn et al. worked in HEK293T with transfected constructs and in t(4;11) material; they did not test the other KMT2A partners, so the word 'specific' in their title is a statement about where the mechanism was demonstrated, not a measured exclusion. This entry supplies the exclusion they did not test, in a different assay (oncoprotein occupancy rather than IRX1 binding), and the agreement between the two is therefore convergent rather than confirmatory.
Evidence that argues against this (1)
- Kuhn's mechanism runs in the opposite causal direction from the one measured here: IRX1 acting ON the MLL-AF4 complex to repress HOXA, rather than the oncoprotein acting on IRX1. Both are compatible with the anti-correlation and this analysis cannot distinguish them. Kuhn A et al. 2016
- E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
IRX1 is normally expressed in pro-B cells and megakaryocyte-erythroid progenitors and is aberrantly active in subsets of AML; IRX1 is also inactivated in 16.2% of Down-syndrome myeloid leukaemia, where its loss upregulates the MYC pathway.
The Role of IRX Homeobox Genes in Hematopoietic Progenitors and Leukemia · Nagel S · Genes (Basel) 14:297 2023What this evidence cannot show
A review plus a large sequencing study, both about transcript-level and copy-number events, neither about oncoprotein occupancy. They are cited here to bound what this entry's negative can mean, not to support it. Nagel's own group and Kuhn/Symeonidou disagree about whether IRX1 is expressed in normal pro-B cells; that dispute is recorded in `dispute-is-irx1-expressed-in-normal-b-cell-development` and is NOT resolved here.
Evidence that argues against this (1)
- In Down-syndrome myeloid leukaemia IRX1 is a LOSS-of-function target, inactivated in 16.2% of cases, which is the opposite polarity to the gain-of-programme reading used for iALL-IRX1. Any model in which IRX1 is simply oncogenic has to account for this. Sato T et al. 2024
suspends the reciprocal-phenotype test without touching the 0-of-18 cohort bound
withdraws the promoter-scale AFF1-only pattern as independent support; the gene-scale result is untouched and uncalibrated
IRX1 is absent from the lineage-matched class; its three samples span two lineages
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
expression enrichment in AFF1 by a different assay in a different cohort; but it also shows the occupancy zero is not an expression zero
the pooled artefact pointed toward that entry's reading; the stratified result does not
bounds the scope of the 0-of-18: it covers B-ALL, T-ALL and non-megakaryoblastic AML only
the EPS15 arm of that 0-of-18 is one patient, and this is the partner where the two assays disagree
- ← related toIRX2 is not fusion-occupied at promoter scale in GSE252378: 0 of 27 samples, against a null of 2.73
reuses q147's floor rule verbatim and extends it from IRX1 to IRX2 at promoter scale
q147's rule has no recorded null; at promoter scale it passes a median 10.75% of all promoters
promoter-scale 2 of 8 beside q147's gene-body-scale 3 of 8; a window-size difference, not a contradiction
the authors' own SEACR + Bayesian N/C pipeline returns the identical IRX1 sample set {p279, p318, p454}, so the call is not an artefact of this project's 2.0x rule
the 0-of-18 is a real measurement but IRX1's own pattern carries p = 0.077 at patient level and 0.119 institution-stratified, so it cannot on its own establish that the address requires AFF1
the gene-scale calls in that entry are produced by the 2.0x gate alone; the floor clause it names changes none of them
the same streaming pattern reproduces q147's 21-region bigWig table at Spearman 0.9932
the gene-body denominator that entry never had: an arbitrary gene body is expected occupied in 0.164 of 27 samples, so IRX1's 3 of 27 is 18.3x expectation
the headline is a correct measurement of a pattern that is not what chance gives at sample level; it should be restated with its n (two patients, one institution) rather than withdrawn