AP1S1 (sigma1A) — review notes

UniProt P61966, 158 aa, AP1S1_HUMAN. Accession asserted against the fetched record:
ID AP1S1_HUMAN Reviewed; 158 AA. / AC P61966; — this is the
expected protein, not a merged accession returning something else. PANTHER family
PTHR11753 (ADAPTOR COMPLEXES SMALL SUBUNIT FAMILY), the family shared with AP1S2,
AP1S3, AP2S1, AP3S1, AP3S2, AP4S1.

GOA seeded 53 rows: one IBA, no IDA/IMP/IPI on the human gene at all, and a large
block of Reactome TAS and ComplexPortal NAS rows. The whole experimental base for
human AP1S1 in GOA is thin; almost everything interesting about this protein is in
papers GOA has not used.

1. What the protein is

sigma1A is the small subunit of AP-1, the clathrin adaptor that works at the TGN and
endosomes. The GO definition of GO:0030121 names it explicitly, which settles the
complex-membership annotations without argument: "A heterotetrameric AP-type membrane
coat adaptor complex that consists of beta1, gamma, mu1 and sigma1 subunits ... In at
least humans, the AP-1 complex can be heterogeneric due to the existence of multiple
subunit isoforms encoded by different genes (gamma1 and gamma2, mu1A and mu1B, and
sigma1A, sigma1B and sigma1C)."

UniProt's own FUNCTION is subunit-level and cautious: "Subunit of clathrin-associated
adaptor protein complex 1 that plays a role in protein sorting in the late-Golgi/trans-Golgi
network (TGN) and/or endosomes." Localisation is Golgi apparatus, Cytoplasmic vesicle membrane; Peripheral membrane protein; Cytoplasmic side and Membrane, clathrin-coated pit, all ECO:0000269|PubMed:9733768.

The three sigma1 paralogs are not interchangeable in expression but are in biochemistry:
PMID:20203623 and PMID:27411398. sigma1A is the ubiquitous one; sigma1B is the
neuron-enriched one whose loss causes X-linked intellectual disability.

There is a solved structure containing this exact protein: PDB 4P6Z, 3.0 A, with
S=1-158 mapped to P61966 in the UniProt cross-references — the AP-1 core bound to
HIV-1 Vpu and BST2 PMID:24843023. This matters below.

2. The molecular function: sigma1A is half of the dileucine cargo-signal site

This is the substantive finding of the review and it is entirely absent from GOA.

Cargo sorting signals come in two flavours and they bind different parts of AP-1.
Tyrosine-based YXXØ motifs bind mu1. Dileucine-based [DE]XXXL[LI] motifs bind
neither subunit alone — they bind the gamma-sigma1 hemicomplex:

• PMID:14691137
• PMID:17360967
• PMID:21097499

The "not any single subunit" clause is exactly what the contributes_to qualifier is for,
and it is why the core function below uses contributes_to_molecular_function rather than
molecular_function.

Crucially the evidence is on sigma1A specifically, not on AP-1 generically — which is
the distinction the review is supposed to draw:

• Point mutants of sigma1A itself kill the binding: PMID:21097499
• Mismatched hemicomplexes do not bind: PMID:39269494.
• The structure puts the cargo mimic on sigma1 directly: PMID:24843023. That is the 4P6Z structure, whose
  chain S is this protein.

All three sigma1 paralogs bind these signals — PMID:21097499 — so this is a shared
sigma1 function, not an AP1S1-private one. It is still a function of AP1S1.

There is no GO term for it. GO has GO:0089710 endocytic targeting sequence binding
for the tyrosine-based motif ("a specific peptide sequence, of 4-6 amino acids with an
essential tyrosine (Y) ... which directs internalization by clathrin-coated pits") but no
counterpart for the dileucine motif. Searched QuickGO for "dileucine" (zero hits),
"sorting signal binding", "targeting sequence binding", "leucine motif binding" and the
MF children of GO:0140312 cargo adaptor activity (only GO:0035615). Hence a
proposed_new_terms entry, and an ONTOLOGY knowledge gap. This is a real gap, not an
artefact of one empty search.

On the proposed term's parent: GO:0005515 protein binding is normally a term to avoid, so
the choice was checked rather than defaulted. GO:0089710 — the tyrosine counterpart the
proposal is modelled on — is itself is_a GO:0005515 (is_a ancestors: GO:0005515,
GO:0005488, GO:0003674), so mirroring it puts the two motif-binding siblings in one genus.

The plausible alternative, GO:0005048, fails on parentage: its complete is_a ancestor
set is itself plus GO:0003674, i.e. it hangs directly off the molecular_function root with
no binding ancestor at all — neither GO:0005488 nor GO:0005515 — so it cannot be the genus
of a binding activity.

That term is a label-drift trap and cost a review round, so it is worth recording in full.
Its three naming artefacts disagree with each other:

The two in-repo GO snapshots agree with each other and are both out of date, which makes
them look corroborating when they are one stale reading. This is the same lesson
UNFOLDED_PROTEIN_BINDING.md records for
GO:0140309 — "A relabel is not a redefinition" — reached there from the opposite direction,
where a synonym was promoted to the primary label. Never infer a term's scope from its
label; when a label is contested, decide on the definition and the is_a closure, which is
what the review now does.

3. Bioinformatics: the site is retained, but retention is not the evidence

AP1S1-bioinformatics/ (uv run python dileucine_site.py) fetches nine sigma subunits
live from UniProt, asserts their lengths, and maps the structurally defined sigma2 pocket
onto each by pairwise alignment. Results in
file:human/AP1S1/AP1S1-bioinformatics/RESULTS.md.

Two useful outcomes and one deliberately unflattering one:

1. All five literature-named sigma1A positions verify against the live P61966 sequence:
   R15, A63, V88, I103, and L90 (the MEDNIK variant position).
2. The alignment independently reproduces every homology the paper asserts
   (sigma2 88 -> sigma1A 88, sigma2 103 -> sigma1A 103, sigma2 88 -> sigma3A 94,
   sigma2 103 -> sigma3A 109). The mapping method is therefore sound.
3. The negative control defeats the conservation argument. sigma4 (AP4S1, Q9Y587) is
   in the same PANTHER family and the epsilon-sigma4 hemicomplex demonstrably does not
   bind these signals [PMID:14691137, PMID:21097499], yet it scores 4/5 identical at the
   pocket — the same score as sigma1A, sigma1B and sigma1C.

So "AP1S1 retains the dileucine pocket" is true but carries no weight on its own. The
claim rests on direct sigma1A mutagenesis and on the sigma1-containing structure. I have
recorded the residues as RETAINED/SUBSTITUTED residue_claims anyway, because that is
the checkable form, but the propagation reasoning is not residue-based and says so.

This is worth stating plainly because the mirror error — inferring function from a
retained fold — is as common as inferring loss from a missing residue, and this family
is a case where the fold genuinely does not predict the function.

4. The IBA: one row, a deep pan-AP-sigma node, and the right LCA term

The single IBA is GO:0016192 vesicle-mediated transport, GO_REF:0000033,
WITH/FROM listing 11 donors plus PANTHER:PTN000204281.

I fetched the family PAINT slice (interpro/panther/PTHR11753/PTHR11753-paint.tsv,
already present). PTN000204281 carries two IBD assertions:

Note AP1S1's own accession seeds the GO:0043231 IBD but not the GO:0016192 one, so
this is not a self-referential IBA and there is no question of redundancy either way.
(GO:0043231 does not appear as an IBA row in AP1S1's GOA, so there is nothing to review
for it.)

Every donor resolved (UniProt xref search, plus the GO API for the worm gene):

That is 11 gene-level donors spanning AP-1, AP-2 and AP-3 sigma subunits across fungi,
nematode, fly, rodent and human. The node is the deep pan-AP-sigma ancestor, not an AP-1
node.

This makes GO:0016192 the correct term, not a granularity failure. The tempting
review is "vesicle-mediated transport is vague; AP-1 does TGN-to-endosome sorting, so
MODIFY to something specific". That would be wrong. GRANULARITY_MISMATCH applies only
when the donors agree and a more specific term was available. Here the donors do not
agree on a pathway: AP-1 sigmas work at the TGN/endosome, AP-2 sigmas in endocytosis at
the plasma membrane, AP-3 sigmas in lysosome/LRO delivery. vesicle-mediated transport
is precisely their least common ancestor, and placing a compartment-specific term at that
node would over-propagate it to the AP-2 and AP-3 members. Verdict: NO_FAILURE_CORE.

Supporting detail: the mouse 1:1 ortholog Ap1s1 is itself among the donors, and the term
is genuinely core for AP1S1 (it is what the protein does). I checked the donors' own GO
records via QuickGO; each carries the same IBA plus independent grounding
(e.g. yeast APS1 GO:0006896 IMP PMID:17003107, C. albicans APS3 GO:0006896 IMP
PMID:20870878, pombe vas2 GO:0042147 and GO:0099638 IDA PMID:19624755).

5. The Ensembl Compara rows, and a trap I did not fall into

Two rows come from GO_REF:0000107 with supporting_entities: UniProtKB:P61967 +
ensembl:ENSMUSP00000106709: GO:0043195 terminal bouton and GO:0098793 presynapse.

P61967 is mouse Ap1s1, not AP1S2. I checked (AP1S1_MOUSE, gene Ap1s1,
Mus musculus, Swiss-Prot) rather than assuming the "sigma" donor of a synaptic
annotation must be the neuronal paralog. So these are 1:1 ortholog transfers, the
cleanest kind.

QuickGO shows the mouse source rows are IDA from PMID:20203623, which is titled
"AP-1/sigma1B-adaptin mediates endosomal synaptic vesicle recycling, learning and
memory" — a sigma1B paper. The tempting move is to call this a paralog mix-up and
REMOVE. That would be the exact error CLAUDE.md warns about: the cached record is
abstract-only (full_text_available: false), MGI curators read the full text, and the
paper self-evidently assays both isoforms since its own abstract contrasts them
("The ubiquitously expressed AP-1-sigma1A complex mediates protein sorting between the
trans-Golgi network and early endosomes"). Deferring to the curator. Graded
KEEP_AS_NON_CORE — real, but a neuronal-context localisation, not the core function of
a ubiquitously expressed adaptor subunit.

6. The one annotation that is actually wrong: receptor-mediated endocytosis

GO:0006898 receptor-mediated endocytosis, TAS, PMID:9733768.

The term definition is unambiguous about compartment: "A specific receptor on the cell
surface binds tightly to the extracellular macromolecule (the ligand) that it recognizes;
the plasma-membrane region containing the receptor-ligand complex then undergoes
endocytosis". This is a plasma-membrane process, and AP-1 is not a plasma-membrane
adaptor — AP-2 is. AP-1 recruitment is Arf1-GTP- and PI4P-dependent at the TGN
PMID:15377783 and PMID:23415225. The current statement of AP-1's scope is the same
PMID:39269494.

And the cited paper contains no endocytosis experiment: it is a gamma2-adaptin
identification paper whose sigma1A content is a yeast two-hybrid interaction and whose
localisation result is paranuclear/Golgi. TAS is a traceable author statement, not an
experimental annotation, so the CLAUDE.md protection for IDA/IMP/IPI curator judgment
does not apply, and the argument here is positive and biological rather than
"the abstract is about another gene". REMOVE.

Note the contrast with GO:0005905 clathrin-coated pit, which looks like the same error
and is not. Its definition explicitly extends past the plasma membrane: "Coated pits form
on the plasma membrane, where they are involved in receptor-mediated selective transport
of many proteins and other macromolecules across the cell membrane, in the trans-Golgi
network, and on some endosomes." Decided from the definition, not the label: ACCEPT.

7. Disease and cargo biology

MEDNIK / IDEDNIK syndrome (MIM 609313), autosomal recessive. Originally a founder splice
mutation PMID:19057675, with zebrafish
knockdown rescued by WT but not truncated human AP1S1 mRNA — a clean loss-of-function
demonstration in vivo, in both perturbation directions.

The 2024 paper is the mechanistically decisive one and reframes the disease: the L90P
missense allele, previously thought to cause a distinct non-syndromic enteropathy, causes
full MEDNIK, and it does so by failing at exactly the two sigma1A jobs identified above
PMID:39269494 and PMID:39269494. It also shows sigma1A is required to get AP-1 onto
membranes at all: PMID:39269494 in triple sigma1-KO cells.

Cargo: copper pumps ATP7A/ATP7B PMID:24754424, tight-junction ZO-1 and claudin-3 in
intestinal epithelium PMID:32306098, EGFR recycling versus lysosomal degradation
PMID:37659097. Tyrosinase is a directly demonstrated gamma1-sigma1A dileucine cargo
[PMID:21097499, PMID:39269494], which is the mechanistic link that makes the
ComplexPortal NAS melanosome assembly row defensible rather than a bare complex-level
projection.

sigma1A also has a non-coat role in endosome maturation, but the evidence is mouse
synaptosomes: PMID:27411398. Note a
nuance affinage flattened: Rabex-5 binds sigma1B, not sigma1A, directly — "Unexpectedly,
Rabex-5 binds σ1B, not σ1A" — the sigma1A link is indirect via ArfGAP1. Because this is
mouse-only I have not created a human NEW row for it; it is a knowledge gap instead.

8. Affinage record

Trust gate tripped: self_evaluation_pairwise: tie (not win), flagged in
.affinage.log. Marked LOW_QUALITY in reference_review and every claim I used from
it was re-verified against its PMID. On re-checking, the narrative is substantially
accurate — it found PMID:39269494, PMID:19057675, PMID:32306098, PMID:27411398 and
PMID:37659097, which is a good haul. Two imprecisions: it reports PMID:2040623 as though
the cDNA were human, when that paper sequenced rat AP17 and mouse AP19; and it
compresses the ArfGAP1/Rabex-5 result as though sigma1A bound Rabex-5, which the paper
explicitly denies. Citation list is all numeric PMIDs, no bioRxiv ids.

What affinage missed — and it is the whole molecular-function story: the entire
dileucine literature. PMID:14691137, PMID:17360967 and PMID:21097499 are the three papers
that establish what sigma1A actually does, and none appear in its citation list. Nor
does the structure containing this protein (PMID:24843023 / PDB 4P6Z), nor the AP-1 core
structure (PMID:15377783), nor the Arf1 activation mechanism (PMID:23415225). This is the
documented failure mode exactly: the decisive papers are titled for the complex, the
partner subunit, or a virus, never for AP1S1. Found by searching Europe PMC on
"dileucine signal AP-1 gamma sigma1 hemicomplex" and by resolving PDB 4P6Z through the
RCSB API rather than by searching the gene symbol.

9. Annotation dispositions

• AP-1 complex membership (GO:0030121 x4): ACCEPT. The GO term definition names sigma1A.
• TGN membrane / Golgi membrane / Golgi / cytoplasmic vesicle membrane / clathrin-coated
  pit: ACCEPT — this is where AP-1 works and where sigma1A is required to put it.
• cytosol (Reactome x9): KEEP_AS_NON_CORE. True of the unassembled/uncoated pool; the
  Reactome rows are steps of coat assembly/disassembly cycles, so the cytosolic state is
  half of a real cycle rather than a functional location.
• lysosomal membrane, early endosome: KEEP_AS_NON_CORE — endpoints of AP-1 routes.
• vesicle-mediated transport / protein transport / intracellular protein transport:
  ACCEPT; these are correct and the IBA one is at the right node.
• membrane (HDA): KEEP_AS_NON_CORE, root-ish and the assay cannot say which membrane.
• response to virus (IEP): MARK_AS_OVER_ANNOTATED. One microarray in one cell line,
  AP1S1 not named in the abstract, no functional follow-up.
• receptor-mediated endocytosis: REMOVE (section 6).
• melanosome assembly, platelet dense granule organization, basolateral protein
  secretion: KEEP_AS_NON_CORE — cell-type-restricted, complex-level, NAS.
• terminal bouton, presynapse: KEEP_AS_NON_CORE (section 5).
• clathrin-cargo adaptor activity: ACCEPT, but see the definition caveat — GO:0035615 is
  defined as "responsible for the formation of endocytic vesicles", which does not
  describe AP-1. Raised as a suggested question rather than acted on, since GO_Central
  itself applies this term to AP-1 sigma subunits (pombe vas2 carries it as IC
  contributes_to GO:0030121).
• NEW: one row, GO:0005198 structural molecule activity, IMP, PMID:39269494. GOA gives
  this gene exactly one MF term and it is complex-level and IEA-derived; the subunit's own
  contribution -- holding AP-1 together -- is demonstrated in human cells in both
  directions (wild-type sigma1A co-IPs with gamma1, restores gamma1 levels and restores
  gamma1 membrane staining in triple-sigma1-KO cells; L90P does none of these, dropping
  gamma1 co-IP to 0-13%) and is annotated nowhere. Coded IMP, not IDA, because the readouts
  come from transfected constructs in a knockout background. The more interesting
  subunit-level MF -- dileucine sorting-signal binding -- gets no NEW row because no GO term
  exists for it; that is the proposed_new_terms entry and the ONTOLOGY knowledge gap
  attached to the first core function.

10. Final counts (all scripted, none asserted from memory)

• 53 GOA rows, 53 seeded review rows, reconciled one-to-one on
  (GO id, evidence code, reference, normalized WITH/FROM) by .scratch/reconcile.py.
• Actions: ACCEPT 32, KEEP_AS_NON_CORE 19, REMOVE 1, MARK_AS_OVER_ANNOTATED 1, NEW 1 (54 total).
• 13 rows carry WITH/FROM; all 13 have supporting_entities and all 13 have a
  propagation_review, whose 27 source_entities are generated from those lists rather
  than typed, so they cannot drift.
• 53 references, every one with a reference_review. 175 supporting_text quotes, all
  verbatim per checkquotes.py.
• 4 residue claims, all resolving against the live sequences.