Oncofusion

start / graph / gene-nin

Geneweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q75-dot1l-partners

NIN (ninein) - a centrosomal scaffold recorded as a DOT1L partner under a UniProt isoform that UniProt itself calls dubious

also called ninein · hNinein · KIAA1565 · GSK3B-interacting protein · Glycogen synthase kinase 3 beta-interacting protein

Ninein is a large scaffolding protein that anchors microtubules at the centrosome, the cell's microtubule hub, far from where DOT1L works on chromatin. A single large screen recovered a trace of it in a DOT1L pulldown, and the identifier the database stores for it is a splice variant that UniProt flags as dubious.

The same thing, for a scientist

NIN (Q8N4C6, 2,090 aa) is a core centrosomal scaffold - subdistal appendage, microtubule anchoring, PCM1-dependent - with no nuclear annotation of any kind [E3]. Its sole recorded contact with DOT1L is one spoke of the Hein et al. GFP-DOT1L pulldown: class **B-**, interaction stoichiometry 0.041 (1 NIN per 24 DOT1L), enrichment 1.53 SD, prey-bait profile correlation **0.045**, `core.interactor` not set [E2]. Two things make this row distinctive. First, **the accession IntAct records is `Q8N4C6-10`, and IntAct's own annotation on that row reads *'Dubious isoform produced through aberrant splice sites'*** - it is the leading identifier of a thirteen-member MaxQuant protein group that also contains the canonical Q8N4C6 [E1]. Second, **its -log10 p value of 8.58 is the highest of any prey in the pulldown including the A+ DotCom subunits**, while its enrichment is the second-lowest [E2] - a reproducible measurement of a very small difference, which is exactly the signature of a consistent low-level carry-over. `DOT1L[tiab] AND (NIN OR ninein)[tiab]` and `(MLL OR KMT2A)[tiab] AND ninein[tiab]` both return 0 [E4].

Go deeper
A very significant p value on a very small effect
preyenrichment (SD)-log10 pcorrelationclass
DOT1L (self)9.437.531.000A+
MLLT10 / AF106.923.130.420A+
MLLT6 / AF177.402.200.333A+
MLLT3 / AF93.741.930.275A+
NIN1.538.580.045B-

NIN's p value beats the bait's own. That is not a paradox and it is not a strength: the Welch statistic here measures how reproducible the small enrichment was across three replicates, and a contaminant that carries over at a steady low level is highly reproducible by construction. An enrichment of 1.53 standard deviations that happens identically every time produces a large -log10 p and a near-zero prey-bait correlation at the same time, and NIN has both.

This is the clearest single illustration in this beat of why a p value cannot grade an interaction - it is brief rule 84's shape moved into proteomics. The paper's own composite class, which weights enrichment and correlation alongside the p value, puts NIN at B-, which is the right answer.

The dubious isoform, and what it does to a coverage count

IntAct records this interactor as uniprotkb:Q8N4C6-10, and carries on the same row the annotation isoform-comment: Dubious isoform produced through aberrant splice sites. 1

The reason is mechanical, not a curation error. MaxQuant reports a protein group - here thirteen accessions, Q8N4C6-10;Q8N4C6-7;E9PN67;Q8N4C6-4;C9J066;Q8N4C6-9;Q8N4C6-2;Q8N4C6;Q8N4C6-5;Q8N4C6-6;F8W6D3;Q8N4C6-11;H0YEL6 - being the set of sequences the observed peptides cannot distinguish. The leading identifier is chosen by the search engine's ordering, not by evidence, and isoform 10 happened to lead. IntAct curated the leading identifier, which is standard practice.

What it costs downstream: this project's coverage recheck resolved NIN from the canonical Q8N4C6 and reported the partner as present-but-unwritten, which is right - but an automated pass that wrote the entry from the IntAct accession would have created a gene entry for a dubious splice variant, with a length (2,116 aa in the Hein table) that does not match the canonical protein's 2,090, and nothing in the pipeline would have objected. The general form is in concept-a-protein-group-leading-id-is-not-the-interactor.

Compartment, and the collision check

NIN's UniProt location record is unusually rich and unusually consistent: cytoplasm, cytoskeleton, microtubule organising centre, centrosome and centriole, on six experimental codes, plus a desmosome call by similarity, with notes on tubular arrangement within the centrosome and PCM1-dependence 3. No nuclear annotation, and this is not an under-annotated protein - the pattern this graph's localization-partner-baseline-compartment-is-often-miscalled warns about is sparse annotation hiding a second pool, and NIN's annotation is the opposite of sparse. That makes the compartment argument stronger here than for the other five, though still a prior rather than a proof.

Collision check (brief rule 88), run before the first query: NIN is exactly the shape that collides - three letters, an ordinary English word fragment, and a substring of thousands of words. data/symbol-collisions.json has 62 rows and no row for NIN; the apparent hits on a naive substring grep are the letters nin inside other rows' prose. So no row existed. NIN[tiab] alone is unusable and was never run bare; every query used (NIN OR ninein) inside an AND with DOT1L or MLL/KMT2A, and the querytranslation was read back to confirm PubMed kept NIN as a term rather than expanding it 4. A row for NIN has been proposed in the memo rather than written silently.

Facts and figures
uniprotQ8N4C6
length AA2090
gene symbolNIN
protein nameNinein (hNinein, GSK3B-interacting protein)
uniprot resolutionaccession:Q8N4C6 AND organism_id:9606 AND reviewed:true; one reviewed hit; primary gene name read back as NIN; length 2,090 aa checked
shown for fusionNO - nobody has looked in any fusion. Zero PubMed records pair MLL or KMT2A with ninein.
source database accession
{
  "accession": "Q8N4C6-10",
  "status": "UniProt isoform 10, annotated by IntAct on this row as 'Dubious isoform produced through aberrant splice sites'",
  "length_in_source_table_aa": 2116,
  "why": "leading identifier of a thirteen-member MaxQuant protein group that also contains the canonical Q8N4C6",
  "hazard": "an automated pass writing the entry from the IntAct accession would create a gene entry for a dubious splice variant with a length that does not match the canonical protein"
}
interaction with dot1l
{
  "source_pmid": "26496610",
  "source_class": "proteome-scale AP-MS atlas",
  "intact_interaction_ac": "EBI-11060394",
  "imex_id": "IM-24272-464",
  "detection_method": "psi-mi:MI:0007 anti tag coimmunoprecipitation",
  "interaction_type": "psi-mi:MI:0914 association",
  "expansion": "psi-mi:MI:1060 spoke expansion - binary pair inferred, not observed",
  "intact_miscore": 0.35,
  "author_confidence_class": "B-",
  "core_interactor": false,
  "interaction_stoichiometry_prey_per_bait": 0.04132,
  "interaction_stoichiometry_as_ratio": "1 NIN per 24 DOT1L",
  "log10_prey_bait_ratio": -1.384,
  "abundance_stoichiometry_prey_per_bait": 0.2541,
  "log10_prey_bait_expression_ratio": -0.5950536,
  "enrichment_sd": 1.533,
  "neg_log10_p": 8.57937,
  "neg_log10_p_note": "the highest in the pulldown, exceeding the bait's own 7.529, on the second-lowest enrichment - reproducibility of a very small difference",
  "prey_bait_correlation": 0.045,
  "valid_values": 3,
  "hela_copies_per_cell": 1074.51,
  "independent_sources": 0,
  "reciprocal_tested": false,
  "reciprocal_note": "NIN is not one of the study's 1,330 bait cell lines"
}
localization
{
  "subcellular": "Cytoplasm, cytoskeleton, microtubule organizing center, centrosome and centriole, on six UniProt experimental codes, plus desmosome by similarity. No nuclear annotation. This is a densely annotated protein, so the absence of a nuclear pool is a stronger signal here than in a sparsely annotated record.",
  "source": "UniProt Q8N4C6, fetched 2026-09-21; six ECO:0000269 codes, none of the cited papers opened (brief rule 91)"
}
absence claim
{
  "queries_run": [
    "DOT1L[tiab] AND (NIN OR ninein)[tiab]",
    "(MLL OR KMT2A)[tiab] AND (ninein)[tiab]"
  ],
  "records": [
    0,
    0
  ],
  "operand_controls": {
    "DOT1L[tiab]": 710
  },
  "names_tried": [
    "NIN (only inside an AND)",
    "ninein"
  ],
  "names_not_tried": [
    "hNinein",
    "KIAA1565",
    "GSK3B-interacting protein"
  ],
  "collision_registry_checked": "yes - data/symbol-collisions.json, 62 rows, grepped 2026-09-21 before the first query; NO ROW EXISTS for NIN, and a row is proposed in the beat memo rather than written silently",
  "date": "2026-09-21",
  "tool": "PubMed E-utilities esearch",
  "bounded_by": "titles and abstracts only; supporting figures not searched"
}
How this could bite us back
  • Written down without its grade, 'DOT1L interacts with ninein' invites a centrosome story in a chromatin disease, and nothing in this graph would then be able to trace it back to one B- row.
  • The p value is the trap in this entry. It is the most significant number in the whole pulldown and it is attached to the second-weakest hit, so any future filter built on p values would promote NIN above AF9.
  • If a later pass writes gene-nin from the IntAct accession rather than from this entry, the graph will carry a dubious splice isoform as a named partner of a core protein.
Still open
  1. Is there a mitotic window in which DOT1L and a centrosomal scaffold could meet? The nuclear envelope breaks down in mitosis and DOT1L has documented mitotic chromatin behaviour, so the one compartmentally coherent version of this interaction is cell-cycle-restricted and has never been looked for.
  2. How often does a large -log10 p value ride on a small enrichment across this atlas, and does the paper's confidence class systematically override it? That is answerable from the 31,944-row table already downloaded here.
  3. Which of the thirteen accessions in the NIN protein group were actually distinguishable by the observed peptides? Only an isoform-resolved reanalysis of the deposited data can say, and until it is done this row names a protein rather than a protein product.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The entire recorded interaction evidence for NIN with DOT1L is one row of one IntAct interaction record - EBI-11060394 / IM-24272-464, spoke-expanded (MI:1060) from a single anti-tag co-immunoprecipitation of LAP-tagged DOT1L in HeLa, interaction type 'association' (MI:0914), intact-miscore 0.35, author-confidence B- - from PMID 26496610, and the interactor accession IntAct records is Q8N4C6-10, carrying the annotation 'Dubious isoform produced through aberrant splice sites'.

    What this evidence cannot show

    One database on one day. The isoform-comment is IntAct's own annotation field on that row and reflects UniProt's characterisation of Q8N4C6-10; I did not independently open the UniProt isoform record. The observed peptides cannot distinguish the thirteen group members, so 'the interactor is a dubious isoform' is not a claim that the dubious isoform is the species present - it is a claim about which identifier the database stores.

    Evidence that argues against this (1)
    • Curating the leading protein-group identifier is standard IMEx practice and the canonical Q8N4C6 is inside the same group, so nothing here shows the identification was wrong - only that the stored accession is not the one a consumer would expect. EMBL-EBI IntAct 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    In the source study's interactions table the DOT1L-NIN spoke is class B- with interaction stoichiometry 0.04132 (1 NIN per 24 DOT1L, log10 prey:bait -1.384), abundance stoichiometry 0.254, enrichment 1.533 SD, prey-bait profile correlation 0.045 and core.interactor not set - while its -log10 p value of 8.579 is the highest of any prey in the pulldown, exceeding the bait's own 7.529; NIN is present in the paper's HeLa proteome at 1,075 copies per cell against DOT1L's 3,779.

    What this evidence cannot show

    Downloaded 2026-09-21 from the Elsevier static CDN with magic bytes checked (PK). The interpretation that a large -log10 p on a small enrichment indicates reproducible carry-over is MY reading of a Welch test over three replicates, not a statement from the paper, whose methods are unreadable. The Hein table reports NIN's sequence length as 2,116, which is the leading group member's length and not the canonical protein's 2,090.

    Evidence that argues against this (1)
    • A highly reproducible enrichment is also what a genuine weak interaction looks like, and the paper's own thesis is that the interactome is dominated by weak substoichiometric contacts. Reading reproducibility as evidence of contamination is a choice the data alone do not force. Hein MY et al. 2015
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProtKB Q8N4C6 (NIN_HUMAN) is human ninein, 2,090 aa, 'Ninein (hNinein) / Glycogen synthase kinase 3 beta-interacting protein', gene synonym KIAA1565, with subcellular location cytoplasm / cytoskeleton / microtubule organizing center / centrosome on six experimental codes (ECO:0000269 on PMIDs 11004522, 11956314, 12403812, 12927815, 14654843 and 15190203) plus centriole and desmosome calls, and no nuclear annotation of any kind.

    What this evidence cannot show

    Resolved 2026-09-21 by accession with the primary gene name read back as NIN and the length checked. Per brief rule 91 the six ECO:0000269 codes mean curators read experiments in those papers, not that any of them is the founding or best study; none was opened. The Hein protein group's leading member is a different length (2,116) from this canonical record, which is a reminder that the row and the accession describe different objects.

    Evidence that argues against this (1)
    • Centrosomal proteins are not categorically excluded from chromatin biology - several have documented moonlighting nuclear roles, and ninein's own partner GSK3B is nucleocytoplasmic. A rich centrosomal annotation lowers the prior for a nucleoplasmic contact without excluding one. UniProt Consortium 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Counted absence, PubMed E-utilities esearch run 2026-09-21: DOT1L[tiab] AND (NIN OR ninein)[tiab] returns 0 records and (MLL OR KMT2A)[tiab] AND ninein[tiab] returns 0 records, with the operand control DOT1L[tiab] returning 710; the querytranslation confirmed PubMed kept NIN as a literal term rather than expanding it.

    What this evidence cannot show

    Bounded by titles and abstracts (brief rule 80). NIN is a three-letter symbol of exactly the shape brief rule 88 warns about and was never searched bare - both queries constrain it inside an AND. data/symbol-collisions.json was grepped first, has 62 rows and contains no row for NIN; the apparent grep hits were the letters 'nin' inside other rows' prose. hNinein and KIAA1565 were not run as separate spellings.

    Evidence that argues against this (1)
    • A zero on a three-letter symbol constrained inside an AND can also arise because the AND is too tight rather than because the literature is empty, and no positive control was run for the NIN limb specifically - only for the DOT1L limb. NCBI PubMed (queries run 2026-09-21) 2026
Connected to