AIG1 (Q9NVV5) — review notes

Human AIG1, 238 aa, chromosome 6. HGNC:21607. PANTHER PTHR10989:SF11,
InterPro IPR006838 (ADTRP/AIG1), Pfam PF04750 (Far-17a_AIG1).
TCDB 9.B.203.1.1 "the aig1 lipid hydrolase (aig1) family".

Bottom line

AIG1 is a polytopic membrane threonine hydrolase that cleaves the ester bond of
fatty acid esters of hydroxy fatty acids (FAHFAs). Catalysis depends on Thr43
(nucleophile) and His134 (general base); both residues are present in the UniProt
feature table with ECO:0000269|PubMed:27018888 and both were mutated to Ala with
loss of activity [PMID:27018888 "The FAHFA hydrolase activities of AIG1 and ADTRP were
abolished by mutating their putative catalytic nucleophilic residues Thr-43 and Thr-47,
respectively"; "Mutation of His134 to Ala in AIG1 (H134A) eliminated FP-labeling"].

The gene's name is an expression observation — it was cloned as a
dihydrotestosterone-inducible transcript from dermal papilla cells
PMID:11266118,
and that paper closes "Further study will be needed to understand the functions of AIG1
in the androgen-regulated hair cycle."

Hypotheses from the task brief, and what the data said

Three of the four predicted defect classes did not confirm. Recording the nulls.

brief hypothesis outcome
"the measured hydrolase activity may not have reached GO" NOT CONFIRMED. GO:0120573 FAHFA hydrolase activity exists (created 2026-03-14) and AIG1 holds it twice — IMP from PMID:27018888 and IEA from 12 RHEA reactions. No MF coverage gap.
"the residual GO record is just the expression observation" NOT CONFIRMED. GOA carries no androgen-response, hair-follicle, HCC-biomarker or NFAT term. PMID:11266118 and PMID:21622095 (the biomarker paper) produced no GO annotation at all.
"a domain-derived catalytic MF with no measurement" (the AGFG1 pseudoenzyme shape) NOT CONFIRMED. The activity was measured directly on the human protein, in human cells, and both catalytic residues are present as annotated SITE features. This is the opposite of a fold-derived activity.
"check the paralogue: a term whose only support is the sibling" PARTLY CONFIRMED, but it inverts. The IBA WITH/FROM is PANTHER:PTN001659973 | UniProtKB:Q96IZ2 (ADTRP) | UniProtKB:Q9NVV5 (self). ADTRP is a genuine paralogue co-seed, and it carries its own IMP to both propagated terms from the same paper. So the IBA is not a family-level guess. What is wrong with it is different — see below.

The headline finding: the GO:0016787 IBA is a stale projection

GOA ships GO:0016787 hydrolase activity on AIG1 by IBA (GO_REF:0000033, WITH/FROM
PANTHER:PTN001659973|UniProtKB:Q96IZ2|UniProtKB:Q9NVV5, date 2026-05-28).

PANTHER's own current data say otherwise. Checked against the live authoritative files,
not against GOA's WITH/FROM column:

PTN001659973 GO:0120573 IBD UniProtKB:Q9NVV5|UniProtKB:Q96IZ2 F 20260603 PTN001659973 GO:0042758 IBD UniProtKB:Q96IZ2|UniProtKB:Q9NVV5 P 20251127

Q9NVV5 AIG1 GO:0120573 IBA PANTHER:PTN001659973|UniProtKB:Q9NVV5|UniProtKB:Q96IZ2 F 20260603 Q96IZ2 ADTRP GO:0120573 IBA PANTHER:PTN001659973|UniProtKB:Q9NVV5|UniProtKB:Q96IZ2 F 20260603

So PAINT placed the general term on 2026-05-28 and replaced it with the specific term
six days later, on 2026-06-03. GOA has not ingested that release. The GO:0016787 row
is a snapshot of a node state that no longer exists — SOURCE_STALE_OR_MISSING, not a
curator judgement. Action: MODIFY → GO:0120573.

The repo's own cached slice records the transition, which is how it was noticed:

$ git show 5d1348b100^:interpro/panther/PTHR10989/PTHR10989-paint.tsv
PTHR10989  PTN001659973  GO:0016787  F  IBD  false  ...  20260528
$ git show 5d1348b100:interpro/panther/PTHR10989/PTHR10989-paint.tsv
PTHR10989  PTN001659973  GO:0120573  F  IBD  false  ...  20260603

5d1348b100 = "Refresh and harden PAINT family slices (#2745)", 2026-08-29.

This supersedes a conclusion in the merged ADTRP review

genes/human/ADTRP/ADTRP-ai-review.yaml (merged 2026-07-27, PR #2338) resolves the
byte-identical row as KEEP_AS_NON_CORE and argues:

"PAINT placed this term on 2026-05-28, ten weeks after GO:0120573 was created
(2026-03-14), so the general term is a deliberate judgement and not a stale-term
artefact."

That was correct on the data available on 2026-07-27 — the cached PAINT slice then
still carried GO:0016787 at the node — but it is no longer correct. Two lessons:

  1. The date in the GOA row is the projection date, not evidence about what the node
    currently holds. The node record has to be read from IBD.gaf.
  2. ADTRP's analysis derived the node→term map from GOA's WITH/FROM field
    (analyze_adtrp_propagation.py queries QuickGO with withFrom=PANTHER:<node>), so it
    could only ever recover the terms GOA had already projected. It cross-checked against
    the cached slice, which at the time agreed. Both sources were downstream of the same
    stale release.

ADTRP's identical row should be revisited. Flagged in suggested_questions with both
genes named once, rather than repeated per gene.

The paralogue divergence that is real: plasma membrane

AIG1 and ADTRP both carry GO:0005886 plasma membrane EXP from PMID:27018888, and both
carry the SL-0039 IEA derived from UniProt's SUBCELLULAR LOCATION: Cell membrane line.
The merged ADTRP review already established that PMID:27018888 contains no localisation
experiment, and I re-ran that scan independently on the cached full text
(full_text_available: true) before relying on it. Case-insensitive counts:

absent phrase count present phrase (control) count
plasma membrane 0 membrane fraction 3
cell surface 0 membrane lysates 9
immunofluoresc 0 transmembrane 29
confocal 0 HEK293T 36
subcellular 0 FAHFA 67
localization / localisation 0

The right-hand column is the positive control: the scan finds what is there, so the zeros
are real absences and not a broken grep. These figures reproduce the merged ADTRP review's
numbers exactly, which is the precondition for using them. (They are inflated relative to
the printed paper because the cached file repeats each section, but they are comparable
within the file.) Supplementary figures are not cached, so the scan is scoped to the
cached full text.

What the paper actually shows is recovery of activity in a 100,000 g membrane pellet
PMID:27018888 plus six topology predictors placing the catalytic residues in TM helices.
Its only statement resembling a compartment call is explicitly speculative
["...indicates these enzymes could have evolved to perform hydrolytic chemistry within
the cell membrane environment"].

ADTRP and AIG1 diverge here, and it is the divergence that matters. ADTRP's
plasma-membrane call stands independently on PMID:21868574 (imaging + Triton X-114
partitioning with TFPI and caveolin-1 in endothelial lipid rafts), which is why the ADTRP
review could ACCEPT the term with only an attribution caveat. AIG1 has no such paper.
Its UniProt SUBCELLULAR LOCATION line cites PubMed:27018888 and nothing else, so the
whole plasma-membrane claim on this gene rests on a reference that does not make it.

And the one study that did examine AIG1's localisation places it somewhere else:
PMID:27040980, with the functional readout also ER-based
["AIG1 over-expression slightly increased susceptibility to oxidative stress, which
correlated with an increased ER Ca(2+) concentration in two different cell lines."].

I did not convert that into an ER annotation. full_text_available: false for that
paper, the abstract does not state the species of the topology construct or the identity
of the "two different cell lines", and it is a single study. Both GO:0005886 rows go to
MARK_AS_OVER_ANNOTATED (the protein is genuinely a membrane protein, so the term is not
wrong, merely unsupported on this gene), GO:0016020 membrane is accepted as the
compartment the evidence actually supports, and the PM-vs-ER conflict is recorded as a
knowledge_gaps entry and a suggested experiment.

Note the topology disagreement is independent of the compartment disagreement: UniProt's
FT table gives six TM helices with a cytoplasmic N-terminus, all
ECO:0000255/ECO:0000305|PubMed:27018888 (i.e. prediction, since PMID:27018888 ran no
topology experiment), whereas PMID:27040980 measured five TMs with a luminal
N-terminus by protease protection. The two agree only on the cytosolic C-terminus.

The RCHY1/Pirh2 interaction

GOA's single GO:0005515 row is IPI with UniProtKB:Q96PM5 from PMID:21988832 (the
human liver protein interaction network). Resolving the partner: Q96PM5 is reviewed
Swiss-Prot ZN363_HUMAN, 261 aa, "RING finger and CHY zinc finger domain-containing
protein 1", EC 2.3.2.27 — a RING-type E3 ubiquitin transferase. Canonical length, no
ORFeome-fragment substitution.

Expanding IntAct rather than trusting NbExp=4 (the ACRV1 lesson — a single screen can be
logged as several sub-methods). For this pair the count survives expansion: the four
records from PMID:21988832 are four genuinely different assay types, not sub-methods
of one Y2H:

2 hybrid | pull down | anti tag coip | confocal microscopy      (all EBI-3895963, PMID:21988832)

Contrast the rest of AIG1's IntAct record, which is the pattern the brief warns about:
141 of 161 interactions come from one publication (PMID:32296183) logged as
two hybrid array 47 + two hybrid prey pooling approach 47 + validated two hybrid 47.
None of those reached GOA as GO:0005515 rows, so GOA is already being selective here.

Independent corroboration, 14 years later and in a different laboratory:
PMID:40303337, plus proximity ligation and
truncation mapping. Caveat on institutional independence, stated rather than glossed: the
Y2H papers (PMID:21988832, PMID:21622095) share an author (Huo K) and both are Fudan
University work, so they are not two independent screens; PMID:40303337 is a different
Fudan department (Zhongshan Hospital cardiology) 14 years later with different authors and
different methods.

So the partner is real, and the row is upgraded from bare protein binding to
GO:0031625 ubiquitin protein ligase binding (def: "Binding to a ubiquitin protein ligase
enzyme, any of the E3 proteins"), which the partner's own EC number licenses directly.

What I declined to take from PMID:40303337. It maps the interaction to AIG1 residues
35–93 by deletion. That span crosses TOPO_DOM 31..44 (extracellular), TRANSMEM 45..67
and TOPO_DOM 68..87 (cytoplasmic), i.e. it removes an entire TM helix. A deletion that
large cannot distinguish "this is the contact surface" from "the protein no longer folds
or inserts correctly" — the ACBD3 lesson in deletion form — so no binding-site claim is
made, only that the interaction occurs.

Ambiguity worth flagging rather than resolving. PMID:21622095's abstract reads
"we identified a novel Pirh2-interacting protein, AIG1, by yeast two-hybrid screening and
confirmed its interaction with p53 both in vitro and in vivo." Read literally, the
in vitro/in vivo confirmation is of an AIG1–p53 interaction, not AIG1–Pirh2. The affinage
record paraphrases it as confirmation of the Pirh2 interaction. The abstract is
self-inconsistent and the full text is not cached; nothing in this review rests on that
sentence.

GO:0042758 — the direction trap that wasn't

Initial worry: AIG1 releases free long-chain fatty acids rather than degrading one, so
"long-chain fatty acid catabolic process" might be inverted (the GO:2000738 shape).

Checked instead of assumed. ChEBI classifies the substrate CHEBI:83670 9-PAHSA(1-) as a
long-chain fatty acid anion (OLS4 hierarchical ancestors: fatty acid anion,
long-chain fatty acid anion, lipid). The FAHFA is a long-chain fatty acid, so
hydrolysing it is literally its breakdown, and every UniProt catalytic-activity line gives
PhysiologicalDirection=left-to-right. The term is correct. ACCEPT, matching the merged
ADTRP verdict on the byte-identical row.

Ontology state: GO models FAHFA metabolism only half-way

GO:0120573 FAHFA hydrolase activity was created 2026-03-14 and is the only FAHFA term
in GO — a QuickGO text search for FAHFA returns it and nothing else. Walking the ontology
rather than only searching it (the AHSP rule): GO:1901569 fatty acid derivative catabolic process has five children — fatty acid primary amide, ketone body, icosanoid,
fatty-acyl-CoA and fatty alcohol catabolic processes — and none of them covers FAHFAs. GO
already models the closest analogue, the thioester case, as
GO:0036115 fatty-acyl-CoA catabolic process.

The merged ADTRP review already files the matching proposed_new_terms entry ("fatty acid
ester of hydroxy fatty acid catabolic process", parent GO:0042758). I concur and do
not duplicate the proposal; it is recorded here as an ONTOLOGY knowledge gap so the
AIG1 review is self-contained without filing the same request twice.

Counting and provenance checks

Species discipline on the NEW rows

Every AIG1-specific functional experiment in PMID:38816388 is in mouse — FI3KO/FI3OE
mouse adipocytes, mouse brain and kidney membrane proteomes for ABPP, HFD mice. The paper's
human SGBS adipocyte work is about IRF3, not AIG1; the only hAIG1 strings in the full
text are inside shAIG1. PMID:40303337 is C57BL/6 mice, global and AAV9 cardiac-specific,
with HL-1 (mouse) cardiomyocytes; the HEK293T truncation Co-IPs do not state the species of
the constructs. Both NEW rows therefore take ISS, not IMP/IPI/IDA.

By contrast every experiment behind the existing GO:0120573 and GO:0042758 IMP rows is
human — HEK293T transfection and mutagenesis, shRNA knockdown in LNCaP, and primary human
T-cells — so IMP is the correct code there.

In vivo caveat worth keeping

PMID:32152231 is the in vivo test, and it is careful about which genotype did what:
"Tissues from mice lacking ADTRP (Adtrp-KO), or both AIG1 and ADTRP (DKO) had higher
concentrations of FAHFAs". The single Aig1-KO is not reported as elevating tissue FAHFAs.
So AIG1's individual contribution to whole-tissue FAHFA tone in vivo is not established by
that paper, even though its contribution in human cells is (≈70% of PAHSA hydrolysis lost
on shRNA knockdown in LNCaP). Recorded as a knowledge gap rather than smoothed over.

Sibling-check results, including the nulls

Run because ACRV1/ACRBP established them; reported whether or not they fired.

check result
WITH/FROM resolution 3 tokens, 3 resolved. Q96IZ2 = human ADTRP (reviewed, paralogue co-seed, carries its own IMP to both terms); Q9NVV5 = self (valid, not circular); PTN001659973 = a tree node, not a protein. Zero unresolved.
IntAct NbExp expansion Fired in the benign direction. 4 records = 4 distinct methods for RCHY1, so the count is real. The 141-record HuRI block is the noisy pattern but never reached GOA.
partner accession sanity (TrEMBL / length) Negative. Q96PM5 is reviewed Swiss-Prot at canonical length.
IBA less precise than its donor (ACRV1 shape) Fired. The donors hold GO:0120573 by IMP while GOA's IBA landed on GO:0016787, three levels up — but the cause is a stale GOA ingest, not PAINT's term choice, since PANTHER already projects the specific term.
complex-projection by reference (ACTR8 shape) Negative. PMID:27018888 = 2 entities, per-protein curation.
retraction / erratum One erratum, immaterial (author name).
LCA / heterogeneous-donor caveat Does not apply. Two donors, both characterised in the same paper, both holding the same specific term. No heterogeneous clade is forcing a broad term.
catalytic residues present in the FT table Present. SITE 43 and SITE 134, both ECO:0000269|PubMed:27018888, both with MUTAGEN entries showing loss of activity. The pseudoenzyme defect is absent.

Review round 2 — what the reviewer changed, and the one I was about to get wrong

The PR approved with no blocking items and six suggestions. Four were acted on, two
declined with reasons. The instructive one is #3.

#3, conceded, and it reversed my draft position. The reviewer asked whether
GO:1901800 positive regulation of proteasomal protein catabolic process belongs
alongside GO:0031398. I had drafted a decline on the grounds that the paper's
proteasome statement is hedged — "likely regulated through a ubiquitin-proteasome
mechanism" — before reading the experiments that sentence summarises. They are properly
controlled:

MG132 reverses, 3-MA and bafilomycin A1 do not. That is a route assignment, not an
inference from the ubiquitination result, so the term is not redundant with GO:0031398:
one covers the modification step, the other the catabolic outcome, and each was measured
separately. Added as a third NEW row.

The lesson is the campaign's own and I repeated it: the hedge in an author's summary
sentence is not a measure of the evidence underneath it. I had read this paper
thoroughly for the Pirh2 interaction and skimmed the degradation arm, which is exactly the
ACRBP failure ("cited a paper four times without reading past the abstract") in a milder
form. Declining on a quoted hedge would have looked well-sourced and been wrong.

#2, acted on. Both NEW rows lacked supporting_entities. GO requires a With/From on
a similarity code. Added UniProtKB:Q9D8B1 (AIG1_MOUSE, reviewed) to all three, and
upgraded ISS to ISO: the transfer is from the 1:1 orthologue, not from generic
similarity — Q9D8B1 and Q9NVV5 sit in the same PANTHER subfamily and carry the same
reviewed protein name. Also recorded the reviewer's point that on GO's conventions the
primary annotation belongs on mouse Aig1 with IMP, with the human row as its orthology
projection; neither exists in GOA, so the mouse row is the one a curator should make first.

#4 and #5, acted on together. core_functions[1] gave the Pirh2 arm co-equal billing
with the hydrolase activity on one murine paper plus a Y2H this review itself marks
UNVERIFIED, while the review's own knowledge gap concedes the two functions may not be
separable. The reviewer was right that this argues for secondary placement rather than for
a split. Removed; the claim still lives in three NEW annotation rows with full evidence,
and the knowledge gap now states explicitly why core_functions carries only the
hydrolase. That also dissolves #5 (curation commentary in the description), which existed
only inside the removed entry.

#6, acted on. PMID:21868574 carried the whole AIG1-versus-ADTRP divergence argument
in two reason blocks and a suggested_question without appearing in references, so no
reader could check it. Added with a finding and a reference_review, title copied from the
cached frontmatter rather than written from memory.

#1, declined here and escalated. The merged ADTRP review now asserts, on a
byte-identical row, a verdict this PR disproves — and its supported_by quotes a
node-to-term line that is stale against the repo's own PAINT slice. The reviewer agrees the
scope boundary is debatable. It is a different gene, needing its own history record and its
own validation run, so it belongs in a follow-up PR rather than being smuggled into this
one. Already disclosed in suggested_questions; reported to the campaign coordinator.

Reviewer's own caveat, checked. It could not run just validate (no just/uv in its
sandbox) and relied on this PR's reported result. That result was re-derived after every
edit in this round: ✓ Valid with the one deliberately-unsatisfied warning, checkquotes
57/57 with zero file: quotes, cache_lint exit 0, and terms.csv untouched.

Isoform note, not actioned

Isoforms 5 and 6 truncate at residue 134–138 (VSP_060691 H134→L plus VSP_060692
135–238 missing; VSP_060693/4 similarly), i.e. they destroy or delete the catalytic
His134 and are predicted to be catalytically dead. Isoform 3 (VSP_060690) deletes
48–99. No annotation in GOA carries an isoform qualifier and no experiment has tested
isoform-specific activity, so nothing is asserted; raised as a suggested experiment
instead.