BAIAP2L2/Pinkbar Nucleoplasm (GO:0005654) Hypothesis — Deep Research Report

Gene: BAIAP2L2 (Pinkbar) · Organism: Homo sapiens (NCBITaxon:9606) · UniProt: Q6UXY1 · Ensembl: ENSG00000128298
Focus type: function_assignment · Hypothesis slug: function-hypothesis-go-0005654
Source file: genes/human/BAIAP2L2/BAIAP2L2-ai-review.yaml (free-text selector)

Summary

The seed hypothesis proposes that human BAIAP2L2/Pinkbar localizes to the
nucleoplasm (GO:0005654) and asks whether transient, context-dependent, or
paralog-specific evidence might support a nuclear pool that documented membrane
functions do not exclude. After three iterations examining primary literature, GO/GOA
term-specific records, UniProt, the Human Protein Atlas, paralog annotations, and the
protein sequence itself, the verdict is clear: the nucleoplasm claim is refuted /
over-annotated. There is no experimental, database, or sequence-based source for a
nuclear localization at any level of confidence.

Every direct localization study places BAIAP2L2 at membrane and actin-based
protrusion compartments. The founding characterization localized Pinkbar to
Rab13-positive vesicles and the plasma membrane at intercellular junctions of intestinal
epithelial cells (PMID: 21743456). A wholly
independent body of inner-ear work places it at the tips of shorter-row transducing
stereocilia in cochlear hair cells, where it is a "row-2 complex" component required for
mechanotransduction (PMID: 33151556,
PMID: 34346063,
PMID: 35044843,
PMID: 39093051). This is exactly what its
I-BAR (IMD) + SH3 membrane-curvature architecture predicts, and it is the functional
opposite of a soluble nucleoplasmic protein.

Critically, the hypothesis's proposed escape routes all fail when tested directly.
Term-specific QuickGO queries return zero annotations for GO:0005654 (nucleoplasm),
GO:0005634 (nucleus), and GO:0005730 (nucleolus); neither paralog (BAIAP2, BAIAP2L1)
carries any nuclear annotation; the Human Protein Atlas antibody yields no subcellular
call at any reliability level; and the 529-aa sequence has no classical NLS or basic
patch capable of driving nuclear import. The recommended curation lead is therefore to
not assign GO:0005654 and to flag any existing nucleoplasm annotation for removal,
pending curator verification of its (undocumented) source.

Key Findings

F001 — Nucleoplasm localization is unsupported by all primary literature and current database records

Direct localization experiments across two entirely different tissue systems converge on
membrane compartments and never report a nucleus. In the founding paper, Pinkbar is
"specifically expressed in intestinal epithelial cells, where it localizes to
Rab13-positive vesicles and to the plasma membrane at intercellular junctions"
(PMID: 21743456). These three
compartments — plasma membrane, cell junction, cytoplasmic vesicle membrane — are exactly
what UniProt curates for Q6UXY1, each with experimental evidence code ECO:0000269 tracing
to this paper.

An independent inner-ear literature reaches the identical conclusion in a different cell
type: "the I-BAR protein BAIAP2L2 … localizes to the tips of the shorter transducing
stereocilia in both inner and outer hair cells"
(PMID: 33151556). Stereocilia tips are
apical, actin-based membrane protrusions — again membrane/cytoskeletal, not nuclear.

At the annotation level, QuickGO returns 8 cellular_component annotations for
BAIAP2L2: plasma membrane (IBA/IEA), cytosol (TAS), vesicle membrane (IDA, PMID:21743456),
cytoplasmic vesicle membrane (IEA), cell–cell contact zone (IDA, PMID:21743456), anchoring
junction (IEA), and clathrin complex (IDA, dictyBase, PMID:21401524). None is
GO:0005654 or any nuclear term. The two paralogs behave the same way — BAIAP2 (Q9UQB8)
has 42 CC annotations and BAIAP2L1 (Q9UHR4) has 11 CC annotations, and neither carries a
single nuclear annotation. Targeted PubMed searches for "BAIAP2L2 nuclear/nucleus" and
"I-BAR nuclear localization" returned no relevant papers. The negative result is
convergent across two experimental systems and three databases.

F002 — No GO term-specific, HPA immunofluorescence, or proteomic source supports a nuclear annotation

To exclude a hidden or low-visibility source, each candidate nuclear term was queried
directly. QuickGO term-specific queries for Q6UXY1 return numberOfHits = 0 for
GO:0005654 (nucleoplasm), GO:0005634 (nucleus), and GO:0005730 (nucleolus).
Across all 17 GO annotations spanning all three aspects, no term contains the string
"nucle" — including no NOT-qualified nuclear annotations.

The Human Protein Atlas has a BAIAP2L2 antibody (HPA003043), but its Cell Atlas
Subcellular-location, main-location, and additional-location fields are all null —
BAIAP2L2 has no HPA immunofluorescence localization call at any reliability level. HPA
therefore provides neither a nuclear call nor any competing call. In sum, none of the
three standard curation sources (GO/GOA, UniProt, HPA) offers direct or predicted
nucleoplasm evidence, and the seed claim has no identifiable source document to trace.

F003 — The sequence carries no nuclear-localization-signal signature

A nuclear pool would require a mechanism of import. Sequence analysis of the 529-aa
Q6UXY1 protein found none. The global basic-residue (K+R) fraction is 0.123, and the
two annotated intrinsically disordered regions (residues 221–327 and 403–510) — the most
plausible NLS locations — have K+R fractions of only 0.121 and 0.111, i.e., they are
not basic-enriched relative to the whole protein. This is the reverse of the basic-patch
enrichment characteristic of NLS-bearing nuclear proteins. The longest contiguous K/R
run is 3 residues, below the 4–6 residue basic core of a classical monopartite NLS, and
no strict bipartite NLS motif (KR-x(9–12)-[KR]{3}) is present anywhere in the
sequence. There is thus no sequence-based route for nuclear import, removing the last
plausible rationale for a computational/predicted nuclear annotation.

Mechanistic Model / Interpretation

BAIAP2L2's architecture and every functional study converge on one coherent model: it is
a membrane-deforming I-BAR/IMD-domain protein that senses and shapes plasma-membrane
curvature at actin-based protrusions, working within multiprotein complexes at the
cytoplasm–membrane interface. It has neither a role at, nor a documented route into, the
nucleoplasm.

   BAIAP2L2 / Pinkbar (Q6UXY1, 529 aa)
   ┌───────────────┬──────────────────────────┬────────────┬───────────────────────┐
   │  I-BAR (IMD)  │  disordered 221–327       │ SH3-region │  disordered 403–510    │
   │ membrane      │  (K+R 0.121, no NLS)      │ partner    │  (K+R 0.111, no NLS)   │
   │ curvature     │                           │ binding    │                        │
   └───────┬───────┴──────────────────────────┴─────┬──────┴────────────────────────┘
   │                                          │
   ▼                                          ▼
   Binds/senses negative                     EPS8L2, TWF2, CAPZB2, CIB2,
   membrane curvature; makes                 EPS8, MYO15A, Rho-GTPase Rif
   planar membrane sheets                            │
   │                                          │
   ▼                                          ▼
   ┌──────────────────────────────┐   ┌──────────────────────────────┐
   │ Intestinal epithelium:        │   │ Cochlear hair cells:          │
   │ Rab13+ vesicles + plasma      │   │ tips of shorter (row-2)       │
   │ membrane at cell junctions    │   │ transducing stereocilia       │
   │ (PMID:21743456)               │   │ (PMID:33151556, 34346063)     │
   └──────────────────────────────┘   └──────────────────────────────┘

   NUCLEUS / NUCLEOPLASM (GO:0005654):
       ✗ 0 GO annotations   ✗ no HPA IF call   ✗ no NLS   ✗ no literature report

The two documented contexts — intestinal junctions and stereocilia tips — are different
tissues that share the same molecular logic: membrane-curvature sensing/generation
coupled to actin regulators. In hair cells, BAIAP2L2 is a bona fide "row-2 complex"
component whose stereocilia-tip retention is Ca²⁺- and mechanotransduction-channel
dependent, and it binds EPS8L2, TWF2, CAPZB2 and CIB2
(PMID: 34346063,
PMID: 35044843). None of these partners is
nuclear. The interesting subtlety the seed hypothesis raises — that localization is
context-dependent — is genuinely true (PMID:35044843 shows conditional, Ca²⁺-dependent
targeting), but the conditional target is a membrane protrusion tip, not the nucleus.
The immediate molecular function under test is a CC (localization) assignment, and
every attempt to reconcile it with a nucleoplasm pool fails.

Evidence Base / Evidence Matrix

GO Decision Table (computed provenance, this run)

GO Curation Implications

• Lead (requires curator verification): Do not assign GO:0005654 (nucleoplasm, CC)
  to BAIAP2L2/Q6UXY1. If a nucleoplasm annotation exists in the working review, flag it for
  removal or downgrade to "not supported / source unverified."
• Aspect: The hypothesis concerns a CC term. The evidence supports no nuclear CC
  term (GO:0005654, GO:0005634, GO:0005730). It should not be generalized to
  GO:0005634 (nucleus) — generalizing an unsourced specific term to its parent merely
  propagates the same unsupported claim.
• Retain / support instead: The well-sourced CC terms are membrane/junction/vesicle
  (GO:0005886, GO:0044291, GO:0030659; PMID:21743456) and, for hair-cell context,
  stereocilium/stereocilium tip (GO:0032420; PMID:33151556, 34346063) — all more
  informative than any generic location.
• Review-worthy carry-over: GO:0071439 (clathrin complex, IDA from dictyBase) deserves
  scrutiny as a cross-organism projection that may not reflect human Pinkbar.
• Nucleoplasm is neither "too broad" nor "too narrow" — it is unsupported, so the
  qualifier question does not apply.

Mechanistic Scope

The immediate molecular function is a subcellular localization (CC) assignment, not a
catalytic or process claim. The direct gene-product activity supported by evidence is
membrane-curvature sensing/generation via the I-BAR (IMD) domain, coupled to actin
regulators through the SH3 and disordered regions. This stabilizes membrane–cytoskeleton
interfaces: planar epithelial junction membranes and the membrane at shorter-row
stereocilia tips (the row-2 complex maintaining mechanotransduction).

The downstream phenotypes — stereocilia maintenance, mechanotransduction current, and
hearing (from knockout mice) — are consequences of that membrane/actin activity in
cytoplasmic/plasma-membrane compartments. A nucleoplasm role would imply a distinct,
undocumented function (chromatin/transcription or nuclear transport) with no supporting
assay, interaction, or phenotype. The nucleoplasm claim is therefore not a downstream
phenotype that could be reconciled with the membrane model — it is a separate localization
assertion with no molecular route (no NLS, no importin partner, no reported nuclear pool)
and cannot be rescued as a "context-specific" facet of the documented function.

Conflicts and Alternatives

The hypothesis anticipates several ways a nuclear call could still be correct; each was
tested:

1. Paralog confusion (BAIAP2 vs BAIAP2L1 vs BAIAP2L2). The most likely source of an
   erroneous annotation, but it does not rescue the claim: all three family members
   lack nuclear annotations (42, 11, 8 CC annotations respectively; zero nuclear). The
   hair-cell literature explicitly distinguishes the paralogs by localization
   (PMID: 39093051), and all are
   membrane/actin-associated.
2. Transient/context-dependent nuclear pool. Localization is context-dependent
   (PMID:35044843), but the conditional target is a stereocilium tip, not the nucleus.
   No IF call, proteomic nuclear-fraction hit, or stimulus-induced translocation is
   reported, and no NLS enables import.
3. Isoform-specific localization. No BAIAP2L2 isoform has been reported nuclear; the
   disordered regions where an isoform-specific NLS might sit are not basic-enriched.
4. Database carry-over / unsourced prediction. The most parsimonious explanation: a
   spurious or predictor-derived nucleoplasm term (deep-learning localization tools often
   default disordered, mildly basic proteins to "nucleoplasm"), a historical/withdrawn HPA
   IF call, or antibody cross-reactivity — none of which is curatable evidence. This is
   precisely the artifact class this review is designed to catch.

No evidence was found that conflicts with the refutation; the conflict runs the other
way — the hypothesis conflicts with the entire evidence base.

Limitations and Knowledge Gaps

1. Negative evidence is not proof of absence. The strongest statements are "zero
   annotations," "null HPA fields," and "no NLS." These make a nuclear pool unlikely and
   unsourced but cannot formally exclude a minor/transient nucleoplasmic fraction no study
   has looked for. Checked: GO term-specific queries, UniProt, HPA, NLS scan. Why it
   matters: the claim should be recorded as "unsupported/no source," not "experimentally
   disproven." Resolution: targeted fractionation or validated-antibody IF.
2. Provenance of the seed nucleoplasm term is unidentified. No document asserting
   nucleoplasm could be traced. A curator should inspect the review YAML's cited
   source/prediction log; its absence from all curation sources itself argues against
   curation.
3. HPA has no call at all rather than an explicit non-nuclear call — the antibody
   (HPA003043) may simply lack a validated Cell Atlas entry.
4. NLS prediction is heuristic. Absence of a classical NLS does not exclude
   non-classical or piggyback import, though no partner-mediated route is documented.
5. No wet-lab data were generated; this is a literature/database/sequence assessment
   across three iterations. Computed tables are embedded here rather than saved as separate
   files because the sandbox working directory was read-only.

Discriminating Tests / Proposed Follow-up Actions

• Trace the annotation source (highest priority, lowest cost). Search the working
  review and its evidence chain for the origin of GO:0005654 (evidence code, reference,
  assigned-by). If it is IEA/prediction or paralog carry-over with no experimental
  backing, the case is closed.
• Validated-antibody IF / endogenous-tag live imaging with nuclear (DAPI/lamin) and
  membrane (junction/stereocilia) markers in intestinal epithelial and cochlear hair
  cells — the single most decisive test.
• Subcellular fractionation (nuclear vs membrane/cytosolic) with immunoblot to detect
  any nuclear pool.
• APEX2/BioID proximity labeling — a nuclear interactome would be required to support
  any nucleoplasm role; membrane/actin partners are expected instead.
• Cross-check cell-map proteomic localization datasets for nuclear-neighbor
  enrichment; none is expected.
• Deletion of disordered basic patches to test any predicted nuclear import — expected
  null.

Curation Leads (require curator verification)

• Action: Reject/remove GO:0005654 (nucleoplasm) for BAIAP2L2; document as
  unsupported/likely computational over-annotation.
• Replacement/retained CC terms: GO:0005886 (plasma membrane), GO:0044291 (cell-cell
  contact zone), GO:0030659 (cytoplasmic vesicle membrane), and — for hair-cell context —
  GO:0032420 (stereocilium)/stereocilium tip.
• Candidate references to verify (exact snippets):
• PMID: 21743456 — "Pinkbar, is
  specifically expressed in intestinal epithelial cells, where it localizes to
  Rab13-positive vesicles and to the plasma membrane at intercellular junctions."
• PMID: 33151556 — "the I-BAR protein
  BAIAP2L2, which localizes to the tips of the shorter transducing stereocilia in both
  inner and outer hair cells."
• Suggested question for submitter: What is the primary source of the nucleoplasm
  assignment (predictor name/version, HPA release, or reference)?
• Suggested experiment: Validated-antibody IF + subcellular fractionation in intestinal
  epithelial and cochlear hair cells.

Final Answer

The hypothesis that human BAIAP2L2/Pinkbar localizes to the nucleoplasm (GO:0005654) is
refuted / over-annotated. All direct localization data place it at the plasma
membrane, intercellular junctions, Rab13⁺ vesicles (PMID:21743456), and cochlear
stereocilia tips (PMID:33151556, 34346063); term-specific QuickGO returns zero nuclear
annotations for BAIAP2L2 and both paralogs; the Human Protein Atlas yields no subcellular
call; and the sequence has no NLS or basic-patch signature — so there is no experimental,
database, or sequence basis for a nucleoplasm annotation, which should not be assigned
and, if present, should be flagged for removal pending source verification.

Artifacts

• OpenScientist final report
• OpenScientist final report