PRSS23 (O95084) — review journal

Why this gene is on the contested-function list

GOA carries GO:0004252 serine-type endopeptidase activity and GO:0006508 proteolysis
for PRSS23. Both come from a single electronic source — InterPro signatures
IPR001254/IPR018114 via GO_REF:0000002, evidence code IEA (confirmed by reading
PRSS23-goa.tsv; there is no IDA/IMP/EXP row for either term anywhere in the file).
UniProt likewise assigns EC=3.4.21.- with no experimental provenance and the
SIMILARITY line "Belongs to the peptidase S1 family" is an ECO:0000305 inference.

So the protease assignment for PRSS23 rests entirely on domain-match, and a 2026 paper
now assays the protein directly and finds no activity.

The 2026 direct-assay paper

Akhtar et al., J Biol Chem 2026 — verified: PMID:41985786, "PRSS23 promotes ovarian
cancer peritoneal dissemination independent of protease activity", J Biol Chem 2026,
DOI 10.1016/j.jbc.2026.111450, PMC13194640, full text cached.

Three independent lines of evidence against catalysis:

Sequence/structure — the zymogen activation switch is gone.
PMID:41985786 and, specifically,
PMID:41985786
This matters because in S1 proteases the triad is necessary but not sufficient — the
switch is what orders the activation-domain loops that build the S1 pocket and oxyanion
hole. The substitutions are not a human idiosyncrasy:
PMID:41985786.
Activity-based probe labelling of conditioned media — negative.
PMID:41985786
This is the strongest form of the negative result: the internal positive control (other
secreted serine hydrolases in the same sample) labels fine.
Chromogenic substrate panel — negative, with the right control.
PMID:41985786
The Ser316Ala control is what distinguishes "no activity" from "contaminating activity",
and it also retires an earlier preliminary report of Z-FR-pNA hydrolysis by the authors'
own group.

Summary statement: PMID:41985786 and the explicit annotation lesson:
PMID:41985786

Independent corroboration that the biology is non-proteolytic. In gastric cancer,
PRSS23 binds eIF4E through the protease-homology domain, and
PMID:41985786 The primary report is PMID:39920289 (Oncogene 2025,
"PRSS23-eIF4E-c-Myc axis promotes gastric tumorigenesis and progression"):
PMID:39920289
Two labs, two tumour types, both find the phenotype survives loss of the catalytic
residue(s).

Decision on the catalytic annotations

GO:0004252 and GO:0006508 → REMOVE. Both are IEA-only, derived from the same
InterPro domain match, and both are now directly contradicted by assay in the only paper
that has looked. CLAUDE.md explicitly permits arguing down an over-propagated electronic
inference on biological grounds; nothing here overrules a curator who read a full text,
because no curator ever made an experimental protease annotation for this gene.

I have not proposed a NOT-qualified annotation. A NOT|enables GO:0004252 would be a
defensible alternative reading of the same evidence and is worth an expert's opinion — it
would record the negative result rather than merely deleting the positive claim. I have
put that in suggested_questions instead of asserting it.

Trafficking and localisation

PMID:41985786
UniProt: signal peptide 1-19, N-glycosylation sequons at Asn93 and Asn207, phosphoserine
109 by FAM20C (the basis of the Reactome GO:0005788 ER-lumen row), SUBCELLULAR LOCATION: Secreted.

GO:0005576 extracellular region (IEA, SubCell mapping) — MODIFY to GO:0005615
"extracellular space"; now directly demonstrated by endogenous tagging, and recovery of
the mature fragment from conditioned media places it in extracellular fluid specifically.
(An earlier draft of this note claimed GO:0005615 was obsolete and kept the parent on
that basis. That was wrong — the term is live; cache/go/terms.csv resolves it and it is
in current use across the repo.)
GO:0005788 ER lumen (TAS, Reactome R-HSA-8952289 "FAM20C phosphorylates FAM20C
substrates") — real but transit-route only; non-core.
GO:0070062 extracellular exosome (HDA, PMID:19199708, parotid-gland exosome MudPIT) —
a bulk-proteomics observation consistent with secretion; non-core.
GO:0005634 nucleus (IDA, GO_REF:0000054, LIFEdb GFP-fusion screen) — flagged as
over-annotated, not removed. It is a single high-throughput fusion-protein screen,
it is not corroborated by any other study, and it is hard to reconcile with a
signal-peptide-bearing protein that endogenous tagging shows entering the secretory
pathway and being secreted. I am deliberately not calling it wrong: I have not seen
the images.

Non-cancer physiology (thin, but real)

Mouse ovary: PMID:18566130
Zebrafish heart-valve development:
PMID:23213106
This is a genuine loss-of-function developmental phenotype, and it is an EndoMT/Snail
signalling role, not a demonstrated proteolytic one — the paper predates the activity
assays and assumed protease function.
The 2006 comparative-genomics paper that seeded the "PRSS23 is an active protease"
framing is explicit about its reasoning being sequence-based:
PMID:16870946 — exactly the triad-only inference the 2026 paper warns about.

Cancer phenotype (the only well-powered functional data)

PMID:41985786
I did not turn this into GO annotations. These are tumour-cell phenotypes in knockdown
lines, not a normal-physiology process the protein carries out, and the mediating molecular
activity is unknown.

What I could not settle

The molecular function of PRSS23 is genuinely unknown. The 2026 paper surveys the
mechanistic options for S1 pseudoproteases — receptor ligand, non-catalytic cofactor/decoy
in a protease cascade, scaffold — without assigning PRSS23 to one. I therefore wrote the
core_functions entry with no molecular_function term at all: a location, a
description and the supporting quotes. Asserting even "protein binding" would be worse than
saying nothing.

The only named direct partner from a focused study is eIF4E (PMID:39920289), and that is
hard to reconcile with a secreted protein — eIF4E is cytosolic. Either there is an
unsecreted intracellular pool, or the interaction is an overexpression artefact. Flagged as
a question, not annotated.

Validation notes

references: titles filled with fill_refs.py; never hand-typed.
Every supporting_text checked as a normalised verbatim substring of the cached
publications/PMID_*.md before validating.