AAGAB (p34, alpha- and gamma-adaptin-binding protein) — review notes

PAINT no-IBA project review, using the affinage deep-research provider
(AAGAB-deep-research-affinage.md, gates passed) plus UniProt Q6PD74, the GOA TSV and the
primary literature.

The problem with this gene's GO record

AAGAB has 18 GOA annotations and not one of them says what the protein does.

• 15 × GO:0005515 protein binding (IPI)
• 3 × cytosol / cytoplasm localisation

There is no molecular function beyond bare binding and no biological process at all.
Meanwhile AAGAB has been the subject of a decade of focused mechanistic work and is, by the
title of one of its own papers, "an assembly chaperone regulating AP1 and AP2 clathrin
adaptors" [PMID:34494650, "AAGAB is an assembly chaperone regulating AP1 and AP2 clathrin
adaptors."]. UniProt is also stale here, offering only a hedged
"May be involved in endocytic recycling of growth factor receptors such as EGFR" from 2012
[file:human/AAGAB/AAGAB-uniprot.txt, "May be involved in endocytic recycling of growth factor"].

This is the largest gap between literature and annotation I have hit in this campaign so far.

What AAGAB actually does

AAGAB is the dedicated assembly chaperone for heterotetrameric AP-type membrane coat adaptor
complexes. It is not a folding chaperone — the subunits are already folded — it enforces an
ordered assembly pathway and protects unassembled intermediates from degradation.

• AP-2. AAGAB guides sequential subunit association; without it, AP2 subunits fail to
  assemble and are degraded PMID:31353312.
• AP-1. AAGAB binds and stabilises the γ and σ subunits; mutation abolishes AP1 assembly
  [PMID:34494650, "AAGAB promotes AP1 assembly by binding and stabilizing the γ and σ subunits
  of AP1"]. Notably it is not required for AP-3
  [PMID:34494650, "However, AAGAB is not involved in the formation of other adaptor complexes,
  including AP3."] — a specificity fact worth preserving.
• AP-4. AAGAB binds and stabilises ε and σ4; knockout cells phenocopy AP-4 subunit mutants,
  accumulating ATG9A at the TGN
  [PMID:35976721, "we report that the alpha- and gamma-adaptin-binding protein (AAGAB, also
  known as p34) binds to and stabilizes the AP-4 ε and σ4 subunits, thus promoting complex
  assembly."]. The same paper states the general point plainly: assembly "is not spontaneous
  but AAGAB-assisted".

Architecture (PMID:36598941): an N-terminal type I pseudoGTPase domain (catalytically
inactive) that engages the small σ subunits, and a C-terminal dimerisation domain that
recognises AP1-γ and AP2-α through a shared surface. AAGAB is a homodimer that converts to
monomer on binding adaptor subunits. PPKP1 disease mutations truncate the CTD, destabilising
the protein and abolishing chaperone function — which ties the molecular mechanism directly to
the human disease. For AP-2 there is a documented handoff to CCDC32 (PMID:39145939), so AAGAB
genuinely does not form part of the finished complex — the exact wording of GO:0051131.

The 15 protein binding annotations are not junk

Every one of them comes from a large-scale interactome screen (HuRI/Rolland, Luck binary
interactome, BioPlex/Huttlin, OpenCell endogenous tagging, Schaffer multimodal cell maps, and
an interactome-perturbation study). My first instinct was to mark them over-annotated. That
would have been wrong.

Resolving the WITH/FROM ids shows what they actually recovered:

AP2S1 is recovered by seven orthogonal methods (Y2H, AP-MS, endogenous tagging,
proximity), and the partner set is precisely the σ and γ/α subunits that the focused
mechanistic literature identifies as AAGAB's clients. These are real, reproducible,
mechanistically meaningful interactions recorded under an uninformative term. The right action
is MODIFY to something informative, not MARK_AS_OVER_ANNOTATED.

HEATR1 is the exception: recovered in only one study, unreplicated by any other screen,
and a nucleolar ribosome-biogenesis protein with no mechanistic connection to adaptor assembly.
(AAGAB-uniprot.txt:171 records NbExp=3 IntAct experiments for the pair, but all three come
from that single report, so it is one study rather than one experiment.) That one is marked
over-annotated.

This produces a deliberate ⚠ WARN about inconsistent actions on GO:0005515 (14 MODIFY,
1 MARK_AS_OVER_ANNOTATED). The inconsistency is real biology — most of these interactions are
the protein's core clients, one is screen noise — so the distinction is kept.

Term choices, and two gaps in GO

• Correction made during review — GO:0035650 does exist. I first recorded that GO has no
  AP-1 adaptor complex binding term, on the strength of an OLS keyword search that returned
  nothing. That was wrong: a direct id lookup confirms GO:0035650 AP-1 adaptor complex binding is a real, non-obsolete term. GO in fact has AP-1 (GO:0035650), AP-2
  (GO:0035612) and AP-3 (GO:0035651) complex-binding terms — and no AP-4 term, which
  is the one adaptor with a genuine gap. The AP-1 entries now use GO:0035650 rather than the
  vague GO:0044877. Lesson: an empty OLS keyword search is not evidence a term is absent —
  confirm by direct id lookup before proposing a new term.
• Scope check on the proposed term's name (added after review round 2). The proposal was
  first drafted as clathrin adaptor complex subunit binding. That name would have excluded
  one of AAGAB's own three clients: AP-4 is not a clathrin adaptor. Verified directly
  against the local GO database rather than assumed —

The term is therefore scoped to GO:0030119, and the same conflation was corrected in the
top-level description and in core_functions.description ("AP-type clathrin adaptor
complexes" → "AP-type membrane coat adaptor complexes").

• Term labels machine-verified. After the GO:0035650 episode, every id used or proposed
  here was resolved against the local go.db via oaklib rather than trusted from memory:
  GO:0035650 = AP-1 adaptor complex binding, GO:0035612 = AP-2 adaptor complex binding,
  GO:0051131 = chaperone-mediated protein complex assembly, GO:0030119/GO:0030131/
  GO:0030124 as tabulated above. Note just fix-labels skips proposed_new_terms and
  proposed_replacement_terms by default (--no-skip-proposed opts in), so proposed-term
  labels are not covered by the routine label check — worth knowing for the rest of this
  campaign.

• The real gap is subunit-level binding. All three existing terms denote binding the
  assembled heterotetramer, whereas AAGAB binds free subunits and is displaced before the
  tetramer exists — PMID:39145939 shows the AAGAB:α:σ2 intermediate "cannot recruit additional
  AP2 subunits" and is handed to CCDC32. A complex-binding term therefore asserts exactly the
  interaction AAGAB's mechanism excludes. GO:0035612/GO:0035650 are used as an explicitly
  flagged interim, with the caveat written into every reason field, and a subunit-binding
  term is filed under proposed_new_terms.

• There is no MF term for assembly-chaperone activity. GO:0044183 protein folding chaperone is explicitly about folding — its definition says "a protein folding chaperone
  binds an unfolded protein to fold it" — and AAGAB's clients are folded subunits awaiting
  assembly. Affinage's grounding proposed exactly this term, and importing it would have been
  a mistake; this is a concrete instance of the AFFINAGE_EVALUATION warning not to import the
  mechanism_profile GO ids. A proper MF term is filed under proposed_new_terms.
• The BP side is well served: GO:0051131 chaperone-mediated protein complex assembly is an
  exact fit ("...mediated by chaperone molecules that do not form part of the finished
  complex"), added as NEW.

Recorded but not annotated

• NEDD4-1 / PTEN / SHIP2 axis in hypoxic-ischaemic injury (PMID:33712741, PMID:41412220):
  rat/cell models, single group, mechanism unclear relative to the adaptor-chaperone role.
• Synaptic vesicle recycling in zebrafish (PMID:38253235): plausible downstream consequence
  of impaired AP-2 assembly rather than a separate function.
• EGFR recycling (PMID:23064416): the original 2012 observation, now best understood as a
  downstream consequence of reduced AP-2. UniProt still leads with this hedge.
• PPKP1 palmoplantar keratoderma: disease association, not a GO process.