PBX2 - an AFF4 co-purifying protein in two rounds of one AP-MS atlas, and the PBX paralogue the leukaemia literature explicitly excluded

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

• The HXR9 peptide this graph carries as target-hox-pbx-interface-peptide-hxr9 disrupts HOX-PBX interactions through the conserved hexapeptide groove, which is shared across PBX1, PBX2 and PBX3. A strategy justified on PBX3 will hit PBX2 as well, and PBX2 is the dominant high-molecular-weight PBX in normal bone marrow - so the on-target normal-tissue exposure of a pan-PBX agent is larger than a PBX3-framed rationale implies.
• If PBX2 does compensate for PBX3, a PBX3-selective intervention has a built-in escape that no published experiment is positioned to see, because no study has depleted more than one PBX paralogue at a time in this disease.
• Equally: nothing in this entry supports treating PBX2 as a target. It is recorded at weak confidence precisely so that a later reader does not upgrade an AP-MS prey into a partner by association with PBX3.

1. Is PBX2 protein present in human KMT2A::AFF1 B-lineage blasts, and at what level relative to PBX1 and PBX3? Selleri's immunoblot says PBX2 dominates the high-molecular-weight PBX pool in murine bone marrow, and no equivalent measurement exists in this disease in either species. A three-lane blot in SEM and RS4;11 would settle it.
2. Does endogenous PBX2 co-immunoprecipitate with endogenous AFF1 or AFF4 in a leukaemia line, or is the BioPlex hit a property of overexpressed tagged ORFs in HEK293T? This is the single experiment that would move the entry off 'weak'.
3. Li 2013 excluded PBX2 on transcript coexpression in AML. Does that exclusion hold at the protein level, and does it hold in B-lineage KMT2A::AFF1 rather than myeloid CA-AML? The graph's whole PBX position rests on a myeloid comparison that has never been repeated in the index lineage.
4. PBX2 is in nuclear speckles in myeloid cells (PMID 10082572, 1999, never replicated) and this graph already holds localization-nuclear-speckles-kmt2a-r. Is PBX2 in speckles in a KMT2A::AFF1 B-lineage blast, and does it move under a menin inhibitor? One immunofluorescence in SEM or RS4;11 would answer a question left open for twenty-seven years.
5. Why does the Blood Atlas call PBX2 undetected in normal immune cells while a 1999 study reports strong nuclear PBX2 protein in myeloid cells? If PBX2 protein is genuinely absent from normal blood and present in transformed myeloid cells, that is a tumour-restricted expression pattern and nobody has said so.

1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   IntAct records PBX2 as a prey of bait AFF4 in two anti-tag coimmunoprecipitation experiments in HEK293T, from BioPlex 2.0 (PMID 28514442) and BioPlex 3.0 (PMID 33961781), both spoke-expanded, MI-score 0.53.

   IntAct molecular interaction database, PSICQUIC MITAB 2.7, query id:P40425 AND id:Q9UHB7, 2 rows retrieved 2026-09-21 · IntAct (EMBL-EBI) 2026
   ▸What this evidence cannot show

   Spoke expansion means neither row is a measured binary interaction; both assert co-purification with the bait. The two rows are the same laboratory, ORF library, cell line and method seven years apart, so they are a reproduction rather than an independent replication. HEK293T is not a haematopoietic cell and the bait is wild-type AFF4, not AFF1 and not a KMT2A fusion.

   ▸Evidence that argues against this (1)

   • An AP-MS prey list from a proteome-scale atlas contains a known background of abundant and sticky proteins, and PBX2 is a nuclear DNA-binding protein of the class that co-purifies with chromatin-associated baits for reasons unrelated to specific contact. Neither BioPlex round reports a reciprocal pulldown with PBX2 as bait. Huttlin EL et al. 2017
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   The BioPlex 3.0 AFF4 pulldown that reports PBX2 also recovers CCNT1, CCNT2, CDK9, ELL, ELL2, ELL3, EAF1, EAF2, HEXIM1 and MLLT1, and PBX2 is a modest-degree prey with 27 distinct partners in BioPlex 3.0 and 16 in HuRI.

   Dual proteome-scale networks reveal cell-specific remodeling of the human interactome. · Huttlin EL, Bruckner RJ, Navarrete-Perea J, et al., Gygi SP · Cell 2021
   PMID 33961781DOI
   ▸What this evidence cannot show

   Partner counts were computed from IntAct's holdings of each publication on 2026-09-21, not from the papers' own supplementary tables, and IntAct's import may not be complete. Recovering the SEC validates the bait, not the individual prey: a pulldown can be simultaneously correct about its core complex and wrong about a peripheral hit. The degree figures bound promiscuity within these datasets only.

   ▸Evidence that argues against this (1)

   • A low prey degree is weak reassurance. A protein can be a specific contaminant of one bait class - here, chromatin- and elongation-associated baits - and still have few partners overall, which is indistinguishable from specificity in a dataset of this design. Rolland T et al. 2014
3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   No focused literature joins PBX2 to AFF4 or to KMT2A: (AFF4[tiab] OR AF4/FMR2[tiab]) AND (PBX[tiab] OR PBX2[tiab]) returns 0 records, and PBX2[tiab] AND (MLL[tiab] OR KMT2A[tiab]) returns 3, all of which title-check to PBX3/HOXA9/MEIS1 papers in which PBX2 appears only as a comparator.

   PubMed counted-absence searches for PBX2 against AFF4, MLL and KMT2A, run 2026-09-21 · PubMed (NCBI) 2026
   ▸What this evidence cannot show

   Operand controls were run and the instrument is working: PBX2[tiab] alone returns 171, AFF4[tiab] alone returns 121, PBX2[tiab] AND leukemia OR leukaemia returns 28, PBX2[tiab] AND (HOXA9[tiab] OR MEIS1[tiab]) returns 16. Both querytranslations expanded as written. The three MLL/KMT2A records are PMID 23264595 (PBX3 is an important cofactor of HOXA9 in leukemogenesis), 22902925 (DACH1 and HOXA9 in MLL-AF9) and 17942707 (Meis1 in MLL leukaemia stem cells). Zero in titles and abstracts; supporting figures were not searched, and a complex-level phrase query such as 'super elongation complex' AND PBX was not run.

   ▸Evidence that argues against this (1)

   • Brief rule 80 is the live risk here: the measurement that would matter - a PBX2 blot in a PBX-family panel from a KMT2A-rearranged cell - is exactly the kind of supporting figure that a title/abstract index cannot reach. The count bounds what is findable by symbol, not what has been done. Li Z et al. 2013
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In cytogenetically abnormal AML including MLL-rearranged AML, it is PBX3 and not PBX1 or PBX2 that is consistently coexpressed with HOXA9, and depletion of Pbx3 inhibits MLL-fusion-mediated transformation.

   PBX3 is an important cofactor of HOXA9 in leukemogenesis. · Li Z, Zhang Z, Li Y, et al., Chen J · Blood 2013
   PMID 23264595DOI
   ▸What this evidence cannot show

   Read as abstract, not full text. The system is cytogenetically abnormal AML - myeloid, not the B-lineage infant KMT2A::AFF1 disease this graph is about - and the transformation assays are murine. The exclusion of PBX2 is a transcript coexpression argument; it is not a protein measurement and it is not a functional test of PBX2. This is the paper that grounds gene-pbx3 and it is cited in 5 entries here.

   ▸Evidence that argues against this (1)

   • A coexpression argument cannot exclude a protein that is present without being transcriptionally co-regulated. Selleri et al. 2004 report by immunoblot that Pbx2 is the predominant high-molecular-weight PBX protein in postnatal bone marrow and thymus, which is not something a coexpression analysis in AML transcriptomes would surface. Selleri L et al. 2004
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In mice, Pbx2-null animals are born at expected Mendelian frequencies with no detectable developmental abnormality, no reduction in long-term survival and no measurable effect on steady-state haematopoiesis or immune function; nevertheless Pbx2 is the predominant high-molecular-weight-isoform Pbx protein detectable by immunoblot in postnatal bone marrow and thymus.

   The TALE homeodomain protein Pbx2 is not essential for development and long-term survival. · Selleri L, DiMartino J, van Deursen J, et al., Cleary ML · Mol Cell Biol 2004
   PMID 15169896DOI
   ▸What this evidence cannot show

   Murine, germline, and steady-state. Read as abstract, not full text. The authors attribute the absent phenotype to redundancy with low-molecular-weight Pbx isoforms, so this is evidence of compensation rather than of irrelevance, and a germline knockout says nothing about a requirement in a transformed cell. No stress haematopoiesis or leukaemia challenge is reported in the abstract. New to this graph - not previously cited here.

   ▸Evidence that argues against this (1)

   • A gene dispensable at steady state can be required under oncogenic stress, and this graph has the worked example: PSIP1/LEDGF is dispensable in normal haematopoiesis and required in the leukaemic cell. A normal knockout phenotype is not an argument that PBX2 is irrelevant to KMT2A-rearranged disease; it is an argument that any requirement would be leukaemia-specific. Oncofusion knowledge graph, mechanism-psip1-required-in-the-leukaemic-cell-not-the-normal-one 2026
6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   PBX2 is UniProt P40425, 430 aa, primary gene name PBX2, sole gene synonym G17, recommended name Pre-B-cell leukemia transcription factor 2; PBX1 (P40424) is also exactly 430 aa and PBX3 (P40426) is 434 aa, so sequence length cannot discriminate PBX1 from PBX2 at all.

   UniProtKB entries P40424 (PBX1_HUMAN), P40425 (PBX2_HUMAN) and P40426 (PBX3_HUMAN) · UniProtKB/Swiss-Prot 2026
   ▸What this evidence cannot show

   Resolved on 2026-09-21 with gene_exact:PBX2 AND organism_id:9606 AND reviewed:true, fields accession,id,gene_primary,gene_synonym,protein_name,length; exactly one reviewed record returned and the primary gene name was read back, then the three lengths were confirmed in a single accession query. gene_exact matches synonyms, which is why the primary-name read-back is the operative check.

   ▸Evidence that argues against this (1)

   • Asserting an accession and a length is a defence against a swap and nothing more. It does not establish that PBX2 protein is present in KMT2A-rearranged blasts, which isoform is present, or that the protein does anything in this disease - and here the length half of that defence is worthless, because PBX1 and PBX2 are identical in length. Li Z et al. 2013
7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   In myeloid cells HOXA9, MEIS1 and PBX2 are all strongly expressed in the nucleus, with a portion of their signals colocalised within nuclear speckles; the same cells yield HOXA9/PBX2/MEIS1 trimeric EMSA complexes and co-immunoprecipitation of PBX2 and MEIS1 with HOXA9 from nuclear extracts.

   HOXA9 forms triple complexes with PBX2 and MEIS1 in myeloid cells. · Mol Cell Biol 1999
   PMID 10082572DOI
   ▸What this evidence cannot show

   ABSTRACT-LEVEL ONLY - the Europe PMC fullTextXML route returned HTTP 500 and the PMC HTML route a stub on 2026-09-21, so the cell line, the antibody, the speckle marker used and any quantitation are not recoverable. A portion of their signals is the paper's own hedge and is not a statement that PBX2 is a speckle-resident protein. One paper from 1999, never replicated for PBX2. No KMT2A fusion anywhere in it.

   ▸Evidence that argues against this (1)

   • The Human Protein Atlas Blood Atlas records PBX2 RNA as Not detected in immune cells with distribution Not detected, which sits directly against strong nuclear PBX2 protein in myeloid cells. Different register (RNA in sorted normal blood cells versus protein in transformed myeloid cells), but unreconciled. Human Protein Atlas 2026
8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   UniProt P40425 asserts PBX2's nuclear location on ECO:0000305 (curator inference) with no experimental citation, while HPA immunofluorescence gives main location Nucleoplasm with reliability Supported.

   UniProtKB P40425 (PBX2_HUMAN) SUBCELLULAR LOCATION comment, fetched 2026-09-21 · UniProt Consortium 2026
   ▸What this evidence cannot show

   An ECO:0000305 location is a curator's inference, not a measurement - the same grade Q71 flagged on SMARCD2. Anyone filtering UniProt on ECO:0000269 would drop PBX2's only location line entirely and record the protein as unlocalised, which would be wrong in the other direction.

   ▸Evidence that argues against this (1)

   • The ECO:0000305 grade impugns the RECORD, not the fact, and it would be an error to downgrade PBX2's nuclear localisation because of it. An entirely independent measurement - antibody immunofluorescence in HPA's panel - returns Nucleoplasm as the single main location at 'Supported' reliability, and a 1999 primary paper reports strong nuclear PBX2 protein in myeloid cells. Three sources of different kinds agree; only UniProt's provenance is thin. The same HPA record is simultaneously the source of this entry's unreconciled Blood Atlas contradiction, so it is being cited here for the arm that supports and elsewhere for the arm that conflicts. Human Protein Atlas 2026

• binds →AFF4 — the textbook SEC scaffold that KMT2A::AFF1 cells need LESS than other leukemias

  co-purifies with tagged AFF4 in BioPlex 2.0 and 3.0 (HEK293T, anti-tag co-IP, spoke-expanded); not a measured binary contact

• related to →PBX3 - the TALE cofactor the field says is non-redundant, and the experiment that says HOXA9 is the redundant one

  PBC-class paralogue; P40425 and P40426, and PBX2 is exactly the length of PBX1 so verify-accessions cannot separate them

• related to →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it

  explicitly excluded as a HOXA9 co-expressed cofactor in MLL-rearranged AML by the paper that established PBX3

• related to →The HOX-PBX interface, and its one peptide: HXR9/HTL001 has been tried in eight tumour types and in KMT2A-rearranged leukaemia exactly once

  the hexapeptide groove HXR9 disrupts is shared across PBX1/2/3, so a PBX3-framed rationale understates the on-target normal-marrow exposure

• related to →How HOXA9 binds DNA: a hexapeptide into a PBX pocket, a 1.9 A structure that is mouse HoxA9 plus human Pbx1 homeodomains, and a second route that makes the motif redundant

  third PBC-class member of the HOX-PBX dimer, previously unwritten in this graph

• ← related toZero of eight unwritten AFF4/MEN1 partners has an experimentally coded nuclear localisation, and the two that are called nuclear are called it by inference

  nuclear by ECO:0000305 curator inference only

• ← depends onThe PBX requirement in KMT2A-rearranged leukaemia is a family property, and every experiment that shaped this graph's view of it removed one paralogue at a time

  the hypothesis is not expressible without a PBX2 entry; gene-pbx3 recorded PBX2 as unresolved

• ← testsBlot all three PBX proteins in KMT2A::AFF1 B-lineage cells, then delete them singly and in combination - the family has never been tested together in any human leukaemia line

  Part A measures PBX2 protein in KMT2A::AFF1 cells for the first time in this disease

• supports →Zero of eight unwritten AFF4/MEN1 partners has an experimentally coded nuclear localisation, and the two that are called nuclear are called it by inference

  one of the two partners with a nuclear call, and the call is curator inference

• related to →Nuclear speckles — the compartment everyone assumes KMT2A::AFF1 uses, on evidence that skips AFF1 itself

  partial nuclear-speckle colocalisation with HOXA9 and MEIS1 in myeloid cells (PMID 10082572); NOT a KMT2A-fusion cell

• binds →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it

  HOXA9/PBX2/MEIS1 trimeric complex from myeloid nuclear extracts, EMSA and co-IP (PMID 10082572)

• binds →MEIS1 — rate-limiting for MLL leukemia stem cells, a surviving co-dependency, and embryonic-lethal when removed

  MEIS1 enhances HOXA9-PBX2 complex formation and joins the trimer (PMID 10082572)

• expressed in →Granulocyte-monocyte progenitor (GMP) — the committed cell an MLL fusion can turn back into a stem cell

  strong nuclear PBX2 protein in myeloid cells (PMID 10082572); contradicted at RNA level by HPA Blood Atlas 'not detected in immune cells'