ABHD8 review notes

What the gene actually does

ABHD8 is named for a fold, and the fold is the source of nearly every problem in its GO
record. Its one characterised function is not catalytic at all — it is a scaffold that brings
an enzyme to a substrate.

The single primary paper is PMID:39225180, which shows ABHD8 binds NLRP3 and routes it into
chaperone-mediated autophagy for degradation: PMID:39225180. The mechanism is
explicitly an adaptor one — the authors' own word is scaffold: PMID:39225180.

Both perturbation directions are reported, which is why the GO:1900226 IMP is solid rather
than suggestive: PMID:39225180 and PMID:39225180. Loss-of-function
plus gain-of-function on the same axis, one in vivo.

The axis is virally targeted: PMID:39225180. UniProt
records this as a separate SUBUNIT line
[file:human/ABHD8/ABHD8-uniprot.txt "CC -!- SUBUNIT: (Microbial infection) Interacts with SARS-CoV-2 nucleoprotein"].

Because the scaffold role is a demonstrated molecular function and no existing annotation
captured it, GO:0140767 enzyme-substrate adaptor activity was added as a NEW IDA. That
accounts for the entry count: GOA has 15 rows and the review has 16 existing_annotations
entries — 15 reviewed rows plus this one proposal, which correctly has no GOA row.

The catalytic annotations, and why none of them is removed

Five IBA rows push lipid-enzyme activities onto ABHD8 (GO:0004620, GO:0052689,
GO:0042171, GO:0006654, GO:0055088), plus an InterPro-derived GO:0003824. No reaction
or substrate has ever been reported for ABHD8. UniProt's own EC number is the maximally
uninformative one — [file:human/ABHD8/ABHD8-uniprot.txt "DE EC=3.-.-.-;"] — which is
the signature of a protein placed in an enzyme family with nothing measured.

All six are MARK_AS_OVER_ANNOTATED rather than REMOVE, and it took a bioinformatics check
to establish that this is the right call rather than a hedge. ABHD8-bioinformatics/
resolves all seven distinct WITH/FROM accessions against UniProt and queries each one's own GO
evidence via QuickGO, at run time
([file:human/ABHD8/ABHD8-bioinformatics/RESULTS.md]).

The single most important result: 6 of the 7 sources carry their own experimental evidence
for at least one propagated term — only the PANTHER node does not. These IBAs are propagating
from genuinely characterised enzymes, not from a family-level guess. That is what makes
over-annotation the right verdict and REMOVE the wrong one — and it is also why the
propagation_review root_cause on every row is PROPAGATION_BAD ("the source annotation
is sound, but the term should not propagate to this target") rather than any SOURCE_* value.
Classifying these as weak sources would now contradict the evidence table.

Evidence provenance and name provenance are different questions, and only one is settled for
all six. The Drosophila source has no reviewed UniProt entry, and its FlyBase id maps to four
UniProt entries carrying different names — Q5U191 "1-acylglycerol-3-phosphate O-acyltransferase
ABHD5" and A1Z753 "Pummelig, isoform A" among them. Its curated GO annotations are real, but
that first name is an automatic by-similarity label. Listing it beside LPAAT_ARATH and CLD1
made an uncharacterised protein look characterised.

Two counts, at two scopes, and they must not be conflated:

scope	sources with own experimental evidence	independently characterised proteins
all five IBA rows	6 of 7 (only the PANTHER node lacks it)	5 of 6 protein sources are reviewed
the two hydrolase rows (`GO:0004620`, `GO:0052689`)	4 of 4 protein sources	3 — the fourth is the unreviewed fly entry

The 3 is what matters for the hydrolase rows' MARK_AS_OVER_ANNOTATED reasoning, since those are
the rows whose sources include the fly entry. Stated globally it would be wrong.

That ambiguity was found only because the resolver was changed to fetch more than one hit and
refuse to pick silently — and the first version of this note then said "two entries" when the
authoritative x-total-results count is four. Inferring a total from a size=2 response is
the very error the check was added to prevent, committed one step further down. The generalisable
point: a size-capped lookup converts an ambiguity
into a confident wrong answer, and it did so here on the one source whose name was doing
argumentative work.

Two source identities were misdescribed in an earlier draft, and both errors ran the same
way — dismissing a source without resolving it:

MGI:MGI:1915938 is not ABHD8's mouse ortholog. It is Q8VD66, mouse ABHD4, a
(lyso)-N-acylphosphatidylethanolamine lipase — a paralog. This is perfectly legitimate for
IBA, because WITH/FROM lists experimentally-annotated members, not orthologs. But it means
no ortholog-strength inference exists anywhere on these rows, and the earlier claim that this
entry "carries the same family-level inference rather than independent experimental support"
was wrong twice over: wrong protein, and it carries IDA.
SGD:S000003342 is CLD1, P53264, yeast mitochondrial cardiolipin-specific deacylase, with
both IDA and IMP. It had been waved away as a distant family member — which is exactly the
mistake already made and corrected for SGD:S000004089/ICT1, repeated on the row next door.

The generalisable rule: an unresolved WITH/FROM accession cannot be dismissed, only deferred.
Every confident sentence about a source needs the lookup behind it, and "this source is just the
same inference" is a testable claim about that source's own evidence codes, not a safe hedge.

The other two facts that decided the verdict:

SGD:S000004089 is a real enzyme, not a mis-transfer — ICT1_YEAST (Q12385), recommended
name 1-acylglycerol-3-phosphate O-acyltransferase ICT1. What is missing is a demonstration
in ABHD8, not an activity in the family.
ABHD8 retains a complete catalytic triad. UniProt annotates three active-site residues
([file:human/ABHD8/ABHD8-uniprot.txt "FT ACT_SITE 252"], plus 370 and 398) and
places it in MEROPS S33
([file:human/ABHD8/ABHD8-uniprot.txt "DR MEROPS; S33.011; -."]). A protein with an intact
charge relay system cannot be called a pseudoenzyme — though note those residues are
themselves ECO:0000250, transferred by similarity.

The defensible position is therefore narrow and slightly unsatisfying: ABHD8's triad is
intact and untested. The fold-derived annotations are premature, not refuted.

The ABHD5 analogy is wrong in both directions — a correction worth recording

My first draft argued these IBAs were fold-only propagation by analogy to ABHD5/CGI-58, "the
textbook α/β-hydrolase fold that is not a hydrolase". That argument was backwards twice over,
and the second error is the more interesting one:

It cuts against my own conclusion. ABHD5 is the fold-without-catalysis example
because it lacks the nucleophilic serine. ABHD8 has the full triad. The comparison, run
properly, is evidence that ABHD8 is more likely to be catalytic, not less.
ABHD5 is not actually a pseudoenzyme. The script's own output shows UniProt recording a
demonstrated CoA-dependent lysophosphatidic acid acyltransferase activity for ABHD5, with
substrate preferences and a primary reference — despite zero annotated active sites. So
ABHD5 has lost the canonical hydrolase serine while retaining a different, measured
chemistry. Invoking it as the family's dead-fold case misstates ABHD5 as well as ABHD8.

Generalising: "protein X has fold F but no activity" is a claim about X's residues, and it
should be checked before it is used as a premise. The fold-name-becomes-activity failure
this review is about has a mirror image — fold-name-becomes-absence-of-activity — and both
are shortcuts around looking at the sequence.

Why the gene is studied at all

Worth recording because it frames the whole record: most ABHD8 literature is not about ABHD8's
activity. It is a candidate target gene of the 19p13.1 breast/ovarian cancer susceptibility
locus — PMID:27601076 (the abstract's ADHD8 is its own typo for ABHD8).

This is cis-regulation of ABHD8 expression, not ABHD8 protein function, so it supports no GO
annotation. But it explains the shape of the evidence: a gene with sustained genetic interest
and exactly one paper on what its protein does. That asymmetry — heavy locus-level attention,
one mechanistic study — is why the fold-derived annotations went unchallenged for so long.

Interaction rows

GO:0005515 appears five times. The two PMID:32296183 rows come from HuRI, a systematic
Y2H map (full text cached), and are marked over-annotated: reproducible but uncharacterised
binary hits, with no follow-up tying either partner to ABHD8's biology. They were initially
merged into one review entry and are now split, one per GOA row, since they concern different
partners. The PMID:39225180 rows are the NLRP3 and ZDHHC12 interactions; one is MODIFY
toward the adaptor MF, the other KEEP_AS_NON_CORE.

GO:0070062 extracellular exosome (HDA, PMID:18570454) is a single neural-stem-cell exosome
proteomics dataset; the paper is abstract-only in the cache. Bulk exosome-proteome membership
is weak evidence of residence for a protein UniProt localises to cytoplasm
([file:human/ABHD8/ABHD8-uniprot.txt "CC -!- SUBCELLULAR LOCATION: Cytoplasm {ECO:0000269|PubMed:39225180}."]),
so it is over-annotated rather than accepted — but not removed, since HDA is a real
observation and I have not read the full text.

Process notes

Recorded here rather than left in PR comments, because these recur.

Never write a reference title: from memory. This was the third occurrence in this
campaign; the pre-write hook hard-blocks it every time, but the wasted round is avoidable.
Copy the title from the cached record: grep -m1 "^title:" -A2 publications/PMID_<id>.md.
propagation_review blocks must not be copy-pasted between rows. GO:0042171's block
was byte-identical to GO:0006654's and consequently omitted AGI_LocusCode:AT4G24160 from
its source_entities. WITH/FROM is per row: the five IBA rows here have 2, 3, 4, 5 and 5
sources respectively. A copied block silently misstates what was inspected.
Diff the working tree against your own claims before pushing. In round 1 I reported
removing the ABHD5 argument from four places (it was removed from three) and splitting the
two PMID:32296183 rows (they were not split). Announcing a fix that did not land costs
more reviewer trust than the original error did — the claim has to be verified against the
tree, not against intent.
file: supporting_text is not machine-checked. The repo validator only verifies PMID:
quotes, so a broken file: quote passes just validate in silence. Every quote in this
review was checked with a local script that applies the same whitespace normalisation
(36 quotes, 0 problems). UniProt quotes in particular must stay inside a single physical
line, or they pick up the CC continuation prefix and stop being verbatim.

2026-09-20 full-gene IBA re-review

Restored broad catalytic/hydrolase inferences and retained lipid homeostasis as non-core. PA biosynthesis/LPAAT remain unresolved due to verified source conflicts. Preserved NLRP3 adaptor function; corrected generic binding and exosome treatment.

No direct target assay alone does not refute an IBD, and an adaptor function can coexist with catalytic functions. Source ABHD5 LPAAT is genuinely contested, not an established function merely missing in ABHD8. The existing bioinformatics report is flagged DISPUTED with exact primary counterevidence. The redundant old NEW adaptor proposal is removed while its row-10 replacement and core function remain.

PMID:39225180: Full primary text establishes ABHD8 recruitment of ZDHHC12 to NLRP3 and absence of the HXXXXD acyltransferase motif. It does not assay and exclude all lipid hydrolase chemistry.
PMID:24879803: Controlled assays attribute CGI-58 LPAAT signal to E. coli PlsC contamination; human CGI-58 expressed in yeast also lacks LPAAT. Contradicts the previous analysis calling human ABHD5 activity established.
PMID:26745266: Reassessment refutes proposed plant CGI-58 acyltransferase and TAG/PC hydrolase activities, while cellular phosphatidylglycerol effects and known homeostasis roles remain. Source activity and source process are not equivalent.
PMID:18570454: Retained curated neural-stem-cell exosome proteomic detection as non-core; no effector role is required for localization.

PAINT: {'family': 'PTHR42886', 'node': 'PTN008676419', 'finding': 'Fetched actual family membership and PAINT slice. Current hydrolase, LPAAT, PA-biosynthesis and homeostasis IBDs were checked. Human ABHD5/yeast Ict1/plant CGI-58 evidence is distinguished from mouse Abhd4/yeast Cld1 hydrolase evidence.'}

All 16 rows were assessed, including experimental, electronic, negated and old proposed entries. All actual GOA rows and source fields remain unchanged. One redundant old reviewer-authored NEW proposal was deleted; the original reviewed-row count includes that proposal. Remaining questions are recorded in projects/IBA_REVIEW/rereview-2026-09-20/receptor-and-lipid-claims.yaml; coordinated reports will be assessed critically when available.