Oncofusion

start / graph / ana-the-dot1l-tcf4-partner-is-tcf7l2-not-tcf4

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q78-interactome-absent

The DOT1L partner IntAct calls TCF4 is TCF7L2 - the second UTP18/WDR5-class symbol mis-assignment found in this interactome, and gene-tcf4 is declined

also called TCF4 TCF7L2 collision · EBI-15894156 · EBI-15894174 · declined gene-tcf4 · IntAct symbol mis-assignment

A database lists TCF4 as a partner of DOT1L. In the paper it comes from, TCF4 is the field's nickname for a different gene, TCF7L2. The two are unrelated proteins, so the entry this project was about to write would have been about the wrong protein.

The same thing, for a scientist

The coverage recheck listed `TCF4` (P15884) among the bait-supported DOT1L partners with no entry. **It should not get one.** Its only records are two spoke-expanded rows from Mahmoudi et al. 2010, *The leukemia-associated Mllt10/Af10-Dot1l are Tcf4/beta-catenin coactivators essential for intestinal homeostasis* [E1][E2]. In the Wnt field **'TCF4' is the standing name for TCF7L2** (Q9NQB0), whose UniProt gene synonym is literally `TCF4` [E3]; the paper's own first sentence is about *'nuclear TCF4/beta-catenin complexes'* in intestinal crypt progenitors, which is TCF7L2 and cannot be anything else. IntAct assigned the bait to **P15884, the class B bHLH factor E2-2/ITF2**, which has no role in Wnt signalling. Three independent internal checks confirm the mis-assignment [E1]. This is the **second** case of its shape in this interactome in two cycles - the first was UTP18 standing in WDR5's slot for twenty-one years - and the two share a mechanism: **a curator resolved a symbol printed in a paper to an accession, and the symbol was the field's nickname for a different gene.**

Go deeper
The three internal checks, all from IntAct's own record

None of these requires reading the paper's figures 1.

  1. The bait recovers itself as a prey. In the mouse arm the bait is Tcf4 (Q60722) and one of its curated preys is Tcf7l2 (Q924A0). If the real bait were Tcf7l2, the bait protein would appear in its own prey list - which is exactly what a mis-assignment produces, and is the same signature as UTP18 sitting in WDR5's slot.
  2. The human arm has no TCF7L2 at all. Q9NQB0 carries 410 IntAct records and not one comes from PMID 21103407, in a paper whose human experiments are in colorectal cancer cells and whose subject is the TCF4/beta-catenin complex.
  3. The prey list is a Wnt/chromatin crypt list. Ctnnb1, Tnik, Carm1, the BAF subunits Smarca5, Smarcc1, Smarcc2, Smarcd2 and Arid1a, Trim28, Chd1l, Wtap, Zc3h13 - a beta-catenin/TCF7L2 crypt complex. E2-2/ITF2's known partners are E-box bHLH heterodimers and are not here.
Why the symbols collide, and why length does not save you
accessionprimary symbolsynonymslengthfamily
Q9NQB0TCF7L2TCF4619HMG box, Wnt effector
P15884TCF4BHLHB19, ITF2, SEF2667class B bHLH (E2-2)
Q924A0Tcf7l2 (mouse)Tcf4459HMG box
Q60722Tcf4 (mouse)Itf2, Sef2670class B bHLH

This is brief rule 63's shape - a curated alias pointing at a different gene - but between unrelated families, which makes it worse than the CSNK2A1/CSNK2A2 case: the wrong answer is not a paralogue whose function might substitute, it is an unrelated protein. Human lengths differ by 48 residues, enough that verify-accessions.py would catch a straight swap, but nothing in this graph would have caught the swap here, because the entry would have been written with P15884 and 667 aa - internally consistent and about the wrong protein.

And the collision compounds with two already logged in this project: gene_exact:TCF3 returns TCF7L1 first, and TCF7L1 carries TCF3 as a synonym. So the TCF7L1/TCF7L2/TCF3/TCF4 quartet contains at least two live symbol traps and data/symbol-collisions.json had a row for none of them when greped on 2026-09-21.

The verdict, and what would reverse it

gene-tcf4 (P15884) is declined. There is no evidence that the class B bHLH factor E2-2 interacts with DOT1L or with any of the six core proteins; the only record attributing it is a mis-assignment, and the records are spoke-expanded on top of that.

gene-tcf7l2 is written instead, carrying the real biology - AF10 and DOT1L as beta-catenin/TCF7L2 coactivators in intestinal crypts, with H3K79 methylation over Wnt target coding regions and an apc-mutant zebrafish rescue.

What would reverse this: any experiment showing E2-2/ITF2 binding DOT1L; or evidence from IntAct's curation record that the Mahmoudi bait was genuinely ITF2. The second is one email to the IntAct curators, and it should go out with the UTP18/WDR5 note the coordinator already has queued.

What this costs the coverage figure: one partner leaves the numerator and the denominator, exactly as UTP18 did last cycle. The pattern is now two for two: every partner in this interactome that turned out to be unwritten because it never existed was a symbol resolved from a paper's text rather than from an accession.

BRINP1: the earlier refusal is upheld

The other declined partner in this beat's set is BRINP1, refused by Q75 in ana-the-six-new-dot1l-partners-are-nine-spokes-of-one-pulldown 4. Re-checked and upheld, on independent grounds. The IntAct record confirms the refusal's premises directly: one row, EBI-11060394, spoke-expanded, interactor recorded as O60477-3 rather than the canonical accession, anti-tag co-IP of LAP-tagged DOT1L in HeLa, from the Hein atlas 5. Adding to the original refusal: on the spoke-expansion census null, BRINP1 is one of only three spoke-only pairs among this beat's thirteen, and it is the only one that is simultaneously spoke-only, isoform-identified, atlas-sourced, class B- and absent from the source study's own proteome of the cell it was measured in. No new entry, and nothing in the original refusal needs amending.

Facts and figures
verdictgene-tcf4 (P15884) DECLINED; gene-tcf7l2 (Q9NQB0) written instead. gene-brinp1 remains DECLINED, upheld on re-check.
classsymbol mis-assignment - a curator resolved a symbol printed in a paper to an accession, and the symbol was the field's nickname for a different gene
precedent in this interactomeUTP18 standing in WDR5's slot in IntAct's curation of PMID 15199122 since 2005-06-03 (ana-utp18-is-not-in-the-mll-complex-it-is-wdr5-miscurated)
n cases now2
effect on coverageone partner leaves the numerator and the denominator, as UTP18 did in cycle 42
outward facing actionan IntAct curator note, to go out with the UTP18/WDR5 note already queued; needs the PI's go-ahead
fetched2026-09-21
internal checks
[
  "the mouse bait Q60722 carries Q924A0 (Tcf7l2) among its own preys - the bait appearing in its own prey list",
  "human TCF7L2 (Q9NQB0, 410 records) has no row from this publication at all",
  "the prey list is a beta-catenin / BAF / crypt list, not an E-box bHLH list"
]
accessions
{
  "declined": {
    "accession": "P15884",
    "symbol": "TCF4",
    "length_aa": 667,
    "family": "class B bHLH (E2-2, ITF2, SEF2)"
  },
  "written_instead": {
    "accession": "Q9NQB0",
    "symbol": "TCF7L2",
    "length_aa": 619,
    "family": "HMG box, Wnt effector",
    "uniprot_synonym": "TCF4"
  }
}
proposed symbol collision rows
[
  "TCF4 / TCF7L2",
  "AR",
  "LXR / NR1H2 / NR1H3",
  "BAT3 / Scythe / G3",
  "WIN (FOXM1 synonym vs the WDR5 WIN motif)",
  "SUMO/ubiquitin-like modifiers as interaction partners"
]
How this could bite us back
  • Declining a partner shrinks a coverage figure that the PI's first priority is measured by. That is the correct direction when the partner does not exist, and the memo says so rather than letting the number drift up.
  • This adjudication rests on internal database evidence and field vocabulary, not on a curator's confirmation. If IntAct replies that the assignment was deliberate, the verdict flips and gene-tcf4 should be written.
  • The second instance of a failure mode is where the temptation to generalise starts. Two cases is two cases; the sentence 'this interactome's curation is unreliable' is not supported by them.
Still open
  1. How many more of this interactome's curated partners were resolved from a paper's printed symbol rather than from an accession? Two cases in two cycles, both found by accident, is not a rate - a systematic pass over the curated partner list against each source's own vocabulary would give one.
  2. Should this graph carry a standing check that flags any curated pair where the bait protein appears in its own prey list? That single signature caught both the UTP18 and the TCF4 cases.
  3. Do TCF7L1 and TCF3 need registry rows too? The graph already logs gene_exact:TCF3 returning TCF7L1 first, and the registry has no row for it.
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct's 42 records from PMID 21103407 assign every transcription-factor row to TCF4/ITF2 (human P15884, mouse Q60722) and spoke-expand all of them; the mouse bait Q60722 carries Q924A0 (Tcf7l2) among its own curated preys; human TCF7L2 (Q9NQB0) has 410 IntAct records and none from this publication; and the curated prey list is a beta-catenin/BAF/crypt list including Ctnnb1, Tnik, Carm1, Smarca5, Smarcc1, Smarcc2, Smarcd2, Arid1a, Trim28, Chd1l, Wtap and Zc3h13.

    What this evidence cannot show

    One database on one day, REST endpoint (brief rule 93). The conclusion that the assignment is wrong is an adjudication from internal evidence plus the paper's own vocabulary; IntAct's underlying curation record was not inspected and the curators have not been contacted. Stated as an adjudication, not as a confirmed database error.

    Evidence that argues against this (1)
    • It is formally possible that the crypt pulldown recovered both ITF2 and TCF7L2 and that the curator assigned each correctly. That would not explain the human arm, where TCF7L2 is absent from a paper about the TCF4/beta-catenin complex in colorectal cancer, but it is the reading that would overturn this adjudication and it is cheap to check with the curators. EMBL-EBI IntAct 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The source publication's subject is Wnt signalling in intestinal crypts - its opening statement is that Wnt maintains crypt progenitors by inducing nuclear TCF4/beta-catenin complexes and that activating Wnt mutations in colorectal cancer drive TCF4/beta-catenin transcription - so the 'Tcf4' whose interactors it reports is the Wnt effector TCF7L2 and not the bHLH factor E2-2.

    PMID 21103407DOI
    What this evidence cannot show

    Abstract read this session; full text not retrieved. The identification rests on field vocabulary, which is a reading of the paper rather than a statement it makes explicitly - the paper never writes 'TCF7L2', because in 2010 nobody in the Wnt field did. That is precisely the trap. Mouse intestine, zebrafish, colorectal lines and HEK293T; no haematopoietic system (brief rule 73).

    Evidence that argues against this (2)
    • The stated reason the trap exists is false, and it is checkable. This caveat says 'the paper never writes TCF7L2, because in 2010 nobody in the Wnt field did'. Measured on PubMed 2026-09-21: TCF7L2[tiab] returns 340 records published up to and including 2010, of which 97 are from 2010 itself - against 95 for TCF4[tiab] in the same year. The current symbol was in routine use when this paper was written, so the authors' choice of 'Tcf4' was a field convention rather than an absence of alternatives, and the adjudication cannot lean on 'nobody used the other name yet'. The adjudication survives on the prey list and the self-as-prey signature; this particular supporting reason does not. NCBI PubMed 2026
    • The field-vocabulary argument has a documented exception: the OTHER TCF4 is itself a Wnt/beta-catenin target. Zhai et al. quantified six candidate beta-catenin/TCF-regulated genes across 44 primary ovarian endometrioid adenocarcinomas and found ITF2 - P15884, the gene whose current official symbol is TCF4 - significantly elevated where beta-catenin regulation was defective, alongside MMP-7, CCND1, CX43 and PPAR-delta. So 'the subject is Wnt, therefore this Tcf4 is the Wnt effector' is not a disjoint classification: both proteins have a place in the Wnt literature of the same tissue, and the argument needs the prey-list evidence to carry it. Zhai Y et al. 2002
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt records Q9NQB0 (TCF7L2, 619 aa, HMG box Wnt effector) with gene synonym TCF4, and P15884 (TCF4, 667 aa, class B basic helix-loop-helix) with synonyms BHLHB19, ITF2 and SEF2, with the same inversion in mouse between Q924A0 and Q60722 - so each gene's official symbol is the other's synonym in the direction that matters.

    What this evidence cannot show

    A statement about identifiers. Primary gene name was read back for every record (brief rules 62 and 65). It does not assert that any existing entry in this graph is wrong - data/symbol-collisions.json was greped for TCF4, TCF7L2 and TCF7L1 on 2026-09-21 and had no row for any of them, so no prior warning existed.

    Evidence that argues against this (1)
    • UniProt's 'synonym' understates it, and HGNC names the relationship precisely: TCF4 is TCF7L2's PREVIOUS OFFICIAL SYMBOL, not an alias. Fetched 2026-09-21, HGNC:11641 TCF7L2 carries prev_symbol ['TCF4'] and only 'TCF-4' as an alias, while HGNC:11634 TCF4 carries no previous symbol at all. That moves the case from brief rule 63's class (a curated alias pointing at a different gene) into rule 69's - a previous official symbol is not an alias - which matters practically: resolvers configured to expand aliases but NOT previous symbols, which is the safe default after the CD15/FUT4 case, will silently fail to connect pre-rename Wnt literature to Q9NQB0 while still connecting 'TCF4' to P15884. The symmetry this claim describes is not symmetric. HGNC 2026
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    BRINP1 was declined as a DOT1L interactor in cycle 42 on five grounds: the prey has no row in the source study's own 8,805-protein HeLa proteome, it is a neural-specific protein, its confidence class is B- with enrichment 1.83 SD and prey-bait correlation 0.084, the recorded interactor is isoform O60477-3 rather than the canonical protein, and PubMed returns zero corroboration under DOT1L, BRINP1, DBCCR1, FAM5A or KMT2A.

    What this evidence cannot show

    This is a citation to an internal analysis entry rather than to a primary source; the primary sources it rests on are PMID 26496610 and its supplementary workbooks, which were not re-read this cycle. The re-check performed here is of the IntAct record only.

    Evidence that argues against this (2)
    • The fifth ground is an ABSENCE claim cited to an internal analysis, which is the one mechanically real defect class the counter-note checker flags - a paper, or an entry, cannot establish what the literature does not contain, and an absence needs a dated bounded query. Re-bounded today: '(BRINP1 OR DBCCR1 OR FAM5A)[tiab] AND (DOT1L OR KMT2A OR MLL)[tiab]' returns 0 on 2026-09-21, with the operand control '(BRINP1 OR DBCCR1 OR FAM5A)[tiab]' returning 76 - so the refusal's absence HOLDS and now has a date and a query behind it. Widening the name list also surfaced a trap worth recording: adding DBC1 raises the count to 2, and both are CCAR2/DBC1 papers (SIRT1-p53 in AML, and an AML methylation profile) - DBC1 is a different gene from DBCCR1 and must not be added to a BRINP1 name list. NCBI PubMed 2026
    • The citation attached to this item resolves, but not to what its title names: the title says 'ana-the-six-new-dot1l-partners-are-nine-spokes-of-one-pulldown, this graph, cycle 42' and the URL is IntAct's findInteractions endpoint for Q8TEK3, which contains no such analysis. A reader following the link gets DOT1L's interaction record, not the argument being cited. The five grounds are correctly reported from that analysis - this is a provenance defect in the pointer, not in the content - but it is the shape that makes internal citations unauditable, and the correct form is either the node id alone or the primary source the analysis rests on (PMID 26496610). EMBL-EBI IntAct 2026
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Re-checked against live IntAct on 2026-09-21, BRINP1's only record against any core bait is EBI-11060394, spoke-expanded (MI:1060), typed MI:0914 association by anti-tag co-immunoprecipitation of LAP-tagged DOT1L in HeLa, with the interactor recorded as O60477-3, from PMID 26496610 - confirming every identifier premise of the cycle-42 refusal.

    What this evidence cannot show

    Confirms the identifiers and the provenance, not the source study's proteome and confidence-class claims, which were not re-derived this cycle.

    Evidence that argues against this (1)
    • BRINP1 is the accession in this beat where the two endpoints diverge MOST, so 'only one record against any core bait' is the most endpoint-dependent statement in the set. Measured the same day, PSICQUIC id:O60477 returns 9 records against IntAct REST's 35 - 26% recovery, against 82% for Q13309/SKP2. And the interactor here is recorded as O60477-3, an isoform identifier: per brief rule 93 a query keyed on the parent accession is not the query that reaches records filed against other isoforms or processed chains, so a completeness claim about BRINP1's core-bait rows is bounded by both the endpoint and the identifier form. The refusal does not depend on completeness - it depends on what the one row is - but the sentence 'only record' does. EMBL-EBI IntAct and PSICQUIC 2026
Connected to