Oncofusion

start / graph / gene-irx2

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q149-irx1-expression-across-partners

IRX2 - the IRX-cluster neighbour that is this disease's own prognostic gene, activated only in KMT2A::AFF1, and never written here until now

also called IRXA2 · Iroquois homeobox 2 · Iroquois-class homeodomain protein IRX-2

IRX2 sits next to IRX1 on chromosome 5 and does a similar job. In infant leukaemia it is switched on only when the fusion partner is AFF1, and the trial that profiled those babies found high IRX2 marked the ones who did worst.

The same thing, for a scientist

🔴 **CORRECTED 2026-09-24 (cycle 138): the MLLT1 count is 1 of 16, not 0 of 16, so IRX2 is NOT confined to KMT2A::AFF1.** q149 reported per-partner IRX2 counts but **persisted no per-sample IRX2 value**, so they could not be checked from its own output. q151 re-derived them from the raw GSE68720 matrix (GPL570 probes `228404_at` and `228462_at`, q149's own max-collapse rule and ten-gene silent floor) and **its instrument control passes**: it reproduces q149's IRX1 numbers exactly (AF4 21/48, named non-AFF1 3/26, germline 3/17). So the IRX2 disagreement is a fact about IRX2 and not about the re-derivation. **`GSM1679755` - a KMT2A::MLLT1 infant, karyotype 47,XY,+X,t(11;19)(q23;p13.3) - has IRX2 at 6.62 against its floor of 6.20, with IRX1 OFF at 3.20.** It is the cohort's **only IRX2-only case**, and it is not an AFF1 case. 🔴 **And the enrichment is weaker than IRX1's, not stronger.** The king re-derived all four Fisher tests exactly: | comparison | table | two-sided p | |---|---|---:| | IRX2: AFF1 vs **all other groups pooled** | 9/48 vs 1/49 | **0.0077** | | IRX2: AFF1 vs **named non-AFF1 partners** | 9/48 vs 1/26 | **0.088** | | IRX2: AFF1 vs germline control | 9/48 vs 0/17 | **0.098** | | **IRX1**: AFF1 vs named non-AFF1 | 21/48 vs 3/26 | **0.0047** | **IRX2 clears 0.05 only when the germline and partner-unknown groups are pooled into the comparator.** Against named non-AFF1 partners alone it does not, and **IRX1 is the stronger partner-enriched signal by an order of magnitude in p.** The coordinator's dispatch brief asserted the opposite - that IRX2 might be the real fusion-associated event and IRX1 the passenger - and that premise is withdrawn. 🔵 **What stands:** the within-AFF1 2x2 below (9 both, 12 IRX1-only, **0 IRX2-only**, 27 both-off) is correct **as scoped to AFF1**. The error was generalising it to *no IRX2-only cases anywhere*, which is a different claim and is false. --- **IRX2** (UniProt **Q9BZI1**, IRX2_HUMAN, **471 aa**, single reviewed human record, gene synonym `IRXA2`) is a TALE-class Iroquois homeobox transcription factor at **5p15.33**, immediately adjacent to **IRX1** (P78414, 480 aa) in the IrxA cluster. This graph names IRX2 in 7 node files and has never written it, while carrying `gene-irx1`, a dispute about IRX1 in normal B-cell development, and a whole IRX1/HOXA subgroup layer. **Two facts make it worth an entry rather than a mention.** **1. It is the prognostic gene in this disease's own infant ALL cohort.** In Children's Oncology Group trial P9407 (97 infant ALL patients), `IRX2` is one of **three genes predictive of event-free survival beyond age and MLL status** - with FLT3 and TACC2 - and differential IRX2/TACC2 expression partitioned the non-low-FLT3 infants into groups with **5-year EFS of 16% versus 64%** (p < 0.001). Within MLL-AFF1 cases specifically, elevated IRX2 sits in a seven-gene classifier correlated with **worse EFS, 20% versus 65%** [E1]. **This graph already cites that paper - for FLT3, in `comut-flt3-kmt2a-r` - and has never used its IRX2 result.** **2. Its activation is confined to KMT2A::AFF1 and is a strict subset of IRX1 activation.** Re-measured from the same deposit with a pre-registered per-sample silent-tissue floor [E2]: | group | n | IRX2 ON | |---|---:|---:| | **KMT2A::AFF1 (`mll_par: AF4`)** | 48 | **9** | | KMT2A::MLLT1 (ENL) | 16 | **0** | | KMT2A::MLLT3 (AF9) | 6 | **0** | | KMT2A::EPS15 | 3 | **0** | | partner unknown | 6 | **0** | | **KMT2A germline infant ALL** | 17 | **0** | **Nine of 48 AFF1 cases, and zero in all 48 other infants.** And all nine are also IRX1-ON: within AFF1 the 2x2 is 9 IRX1+IRX2+, 12 IRX1-only, 27 double-negative, **0 IRX2-only**. So IRX2 marks a sub-subgroup of the iALL-IRX1 state. **Note what is NOT claimed.** Nobody has cross-tabulated the nine IRX2-ON cases against outcome - the EFS variable is not in the GEO record. **The IRX2-high poor-outcome group and the IRX1-high group may be the same children, and that is a hypothesis generated by the nesting, not a result.** Joining two true facts is not a verified fact. **IRX2 and IRX3/IRX5 are not interchangeable.** Nagel's myeloid TALE-code reports **ectopic IRX3 and IRX5** in AML, with IRX1 normally restricted to megakaryocyte-erythroid progenitors [E3]; IRX2 is not part of that headline. In this beat's AML cohort IRX2 has no notable partner pattern.

Go deeper
Why this was invisible, and it is the recurring reason

The paper carrying the IRX2 result, PMID 22210879, was already cited in this corpus - once, in comut-flt3-kmt2a-r, for FLT3. Its IRX2 finding, its TACC2 finding and its GEO deposit (GSE68720) were all unused.

That is this project's own standing rule: search the citations, not the entry text. A paper cited for one of its findings looks covered. Here the paper's other finding was the one that mattered for the IRX1 layer, and the deposit behind it turned out to be the only partner-annotated B-lineage expression cohort in existence.

And the retired-symbol problem compounded it. The deposit is reachable in GEO only through AF4, never KMT2A or AFF1. So the paper was in the bibliography and the data behind it was in a place the corpus's search vocabulary could not reach.

What is not known about IRX2 here

Deliberately short, because the honest answer is almost everything:

  • No occupancy data. GSE252378 has KMT2A-N tracks and this beat did not score IRX2 against them. The q145/q147 promoter interval table covers IRX1; IRX2 was never added.
  • No protein-level data in any haematopoietic context in this corpus.
  • No dependency data. DepMap was UNREACHABLE this cycle (see concept-q149-a-resource-status-has-three-values-and-an-http-200-from-depmap-is-a-bot-wall), so whether IRX2 is required in any KMT2A-r line is NEVER-CHECKED, not absent.
  • No interaction data. finding-irx1-binds-the-mll-af4-complex-without-displacing-menin records a direct physical result for IRX1; nothing analogous exists for IRX2.
  • No normal-development placement in B-lineage. Nagel & Meyer's pro-B TALE-code reports IRX2, IRX3 and IRX5 as silent in normal B-cell differentiation, which is the one thing that is recorded - and it makes the 9 of 48 AFF1 cases an aberrant activation rather than a retained programme 4.
Facts and figures
uniprotQ9BZI1
length AA471
uniprot entry nameIRX2_HUMAN
gene primaryIRX2
protein nameIroquois-class homeodomain protein IRX-2
resolved howUniProt REST gene_exact:IRX2 AND organism_id:9606 AND reviewed:true, fetched 2026-09-24; ONE reviewed record; primary gene name read back as IRX2 and length asserted so verify-accessions.py can catch a swap
length discriminator margin vs IRX19
cytoband5p15.33, IrxA cluster, adjacent to IRX1
gpl570 probes2
epigenetic readoutnone recorded - no histone mark, accessibility or methylation measurement at IRX2 exists in this corpus or was found in this beat
shown for fusionKMT2A::AFF1 only, at the TRANSCRIPT level, in 9 of 48 infant ALL cases; nobody has looked in any fusion at the occupancy or protein level
absence classnever-measured
irx1 irx2 2x2 within AFF1
IRX2_only0.000IRX2_only: 0both_on9.000both_on: 9IRX1_only12.0IRX1_only: 12both_off27.0both_off: 27n48.0n: 48
measurements
count_IRX2_only_cases0.000count_IRX2_only_cases: 0proportion_IRX2_ON0.188proportion_IRX2_ON: 0.1875length_aa471.0length_aa: 471
gene synonyms
[
  "IRXA2"
]
paralogue for contrast
{
  "IRX1": {
    "uniprot": "P78414",
    "length_aa": 480
  }
}
irx2 on by partner GSE68720
{
  "AF4": "9/48",
  "ENL": "1/16",
  "AF9": "0/6",
  "EPS15": "0/3",
  "partner_unknown": "0/6",
  "KMT2A_germline": "0/17",
  "ASAH3": "0/1"
}
irx1 irx2 2x2 within AFF1 — the numbers behind the chart
{
  "both_on": 9,
  "IRX1_only": 12,
  "IRX2_only": 0,
  "both_off": 27,
  "n": 48
}
measurements — the numbers behind the chart
[
  {
    "quantity": "length_aa",
    "subject": "IRX2 human reviewed UniProt Q9BZI1",
    "value": 471,
    "units": "residues",
    "n": 1,
    "universe": "UniProt REST gene_exact:IRX2 AND organism_id:9606 AND reviewed:true, one record",
    "null": 480,
    "note": "the 'null' is IRX1 P78414 at 480 aa - only 9 residues apart, so length is a WEAK discriminator between these two paralogues and gene_primary must be asserted"
  },
  {
    "quantity": "proportion_IRX2_ON",
    "subject": "KMT2A::AFF1 infant ALL",
    "value": 0.1875,
    "units": null,
    "n": 48,
    "universe": "GSE68720 GPL570, max-of-ten-silent-genes per-sample floor, 2 IRX2 probes collapsed by maximum",
    "null": 0,
    "note": "9 of 48; the null is 0 of 49 across all non-AFF1 KMT2A-r and KMT2A-germline infants in the same cohort, rule of three below 6.1%"
  },
  {
    "quantity": "count_IRX2_only_cases",
    "subject": "KMT2A::AFF1 infant ALL, IRX2-ON and IRX1-OFF",
    "value": 0,
    "units": "patients",
    "n": 48,
    "universe": "GSE68720; IRX2-ON is a strict subset of IRX1-ON",
    "null": null,
    "note": "9 double-positive, 12 IRX1-only, 27 double-negative, 0 IRX2-only"
  }
]
corrected cycle 138
{
  "was": {
    "AF4": "9/48",
    "ENL": "0/16",
    "AF9": "0/6",
    "EPS15": "0/3",
    "partner_unknown": "0/6",
    "KMT2A_germline": "0/17"
  },
  "irx2_only_case_cohort_wide": "GSM1679755, KMT2A::MLLT1, IRX2 6.62 vs floor 6.20, IRX1 3.20 OFF",
  "fisher_irx2_aff1_vs_named_non_aff1": 0.088,
  "fisher_irx2_aff1_vs_all_other_pooled": 0.0077,
  "fisher_irx1_aff1_vs_named_non_aff1": 0.0047,
  "note": "IRX2 is not AFF1-confined and its partner enrichment is weaker than IRX1's; it clears 0.05 only against a pooled comparator. The within-AFF1 2x2 is correct as scoped."
}
How this could bite us back
  • IRX2 is a homeobox transcription factor with no chemical matter and a 471-residue paralogue nine residues from IRX1 - so any tool compound or antibody raised against one is a candidate cross-reactor for the other, and the length check cannot catch a swap. Assert gene_primary.
  • If IRX2 marks the worst-outcome KMT2A::AFF1 infants, the temptation is to treat it as a target. It sits inside the IRX1-high arm, which this beat's PROM1 analysis says is the arm least reachable by the disease's one credible surface target - so the IRX2 group is the group with the fewest existing options, not the most.
Still open
  1. Are the 9 IRX2-high KMT2A::AFF1 infants the poor-outcome group of Kang's seven-gene classifier? The nesting says they are inside the IRX1-high group; the EFS variable is not in the GEO record and the cross-tab has never been done.
  2. Is IRX2 fusion-occupied in any KMT2A-rearranged sample? GSE252378 has the tracks and the q145/q147 promoter interval table covers IRX1 and not IRX2 - adding one interval would answer it with no new data.
  3. Is IRX2 required in any KMT2A-rearranged cell line? DepMap was unreachable this cycle, so this is NEVER-CHECKED. But all four CCLE KMT2A::AFF1 lines are IRX1-negative, so a dependency screen in them may be testing the wrong state.
  4. Does IRX2 activation require IRX1, or do they share an upstream switch? Zero IRX2-only cases in 48 is consistent with either, and with IRX2 simply being the weaker readout.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 97 infant ALL cases from COG trial P9407, IRX2 is one of three genes predictive of event-free survival beyond age and MLL status, differential IRX2 and TACC2 expression partitions infants into groups with 5-year EFS of 16% versus 64%, and within MLL-AFF1 cases elevated IRX2 is part of a seven-gene classifier correlated with worse EFS (20% versus 65%).

    PMID 22210879DOI
    What this evidence cannot show

    Read from the abstract, which states the three-gene predictor, the EFS splits and the seven-gene MLL-AFF1 classifier verbatim. The direction of IRX2's contribution within the seven-gene classifier is given as 'IRX2, EPS8 and TPD52 expression were correlated with worse' EFS; the per-gene hazard ratios and the cut-points were not extracted. A classifier derived and evaluated in one 97-patient trial cohort without an external validation set.

    Evidence that argues against this (1)
    • A three-gene and a seven-gene classifier derived from 97 patients on one trial, without external validation, is the classic setting for an over-fitted prognostic signature. IRX2's prognostic value has not been replicated in any independent infant ALL cohort that this corpus has found - and the resource survey in this beat establishes that no second partner-annotated B-lineage cohort exists to replicate it in. Kang H et al. 2012
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In GSE68720, IRX2 transcript exceeds a per-sample ten-gene silent-tissue floor in 9 of 48 KMT2A::AFF1 infant ALL cases and in 0 of 16 MLLT1, 0 of 6 MLLT3, 0 of 3 EPS15, 0 of 6 partner-unknown and 0 of 17 KMT2A-germline cases, and all 9 IRX2-positive cases are also IRX1-positive with no IRX2-only case.

    PMID 22210879DOI
    What this evidence cannot show

    My analysis of their deposited data. IRX2 has two probes on GPL570 collapsed by their maximum. The threshold is the pre-registered max-of-silent-panel floor, which this beat has shown to be threshold-sensitive for IRX1; the IRX2 result is more robust only because the non-AFF1 groups are at exactly zero under it. The all-zero non-AFF1 arms give a rule-of-three bound of 6.1% over the pooled 49 non-AFF1 and germline infants, which is a real bound but rests on the same floor.

    Evidence that argues against this (1)
    • IRX2's cohort median is 4.89 against a per-sample floor typically 6-7, i.e. IRX2 sits closer to the array background than IRX1 does in most samples. So the 9-of-48 is a count of cases far above a background, and the zeros elsewhere could partly reflect that IRX2 simply has less dynamic range on this platform than IRX1. A second platform would be needed to separate those. Kang H et al. 2012
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The myeloid TALE-code reports ectopic expression of IRX3 and IRX5 in acute myeloid leukaemia and places normal IRX1 expression exclusively in megakaryocyte-erythroid progenitors, with IRX2 not part of that headline finding.

    PMID 35328612DOI
    What this evidence cannot show

    DO NOT CITE FROM THE TITLE - it reads as though IRX1 were not aberrant in AML, while the abstract states 'aberrant activity of IRX1 in addition to IRX3 and IRX5'. Cited here for what it says about IRX3/IRX5 versus IRX2, i.e. for a distinction it draws, which is a weaker use than citing a positive finding. Public profiling data plus cell lines.

    Evidence that argues against this (1)
    • The same abstract's cell-line result names 'overexpression of IRX1, IRX3, and IRX5', again omitting IRX2 - so IRX2's absence from the myeloid picture is consistent across both halves of that paper. But absence from a headline is not a measured negative, and the paper may simply not have found IRX2 interesting rather than having found it silent. Nagel S et al. 2022
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The pro-B TALE-code reports IRX2, IRX3 and IRX5 as silent in normal B-cell differentiation, with IRX1 expression placed in pro-B cells.

    PMID 36233173DOI
    What this evidence cannot show

    Public-dataset expression profiling plus cell-line models, not primary sorted human pro-B cells. Its placement of IRX1 in normal pro-B cells is contested and is recorded as dispute-is-irx1-expressed-in-normal-b-cell-development; the IRX2/IRX3/IRX5 silence claim rides on the same instrument and inherits the same doubt.

    Evidence that argues against this (1)
    • If Nagel's normal-pro-B placement is wrong - and Kuhn and Symeonidou treat IRX1 as an aberrant activation, which presupposes it is not the normal programme - then the same survey's claim that IRX2 is silent in normal B-cell differentiation is also suspect, and the 9-of-48 AFF1 IRX2 activations cannot be called aberrant on this basis. Symeonidou V et al. 2021
Connected to