GALNT1 (GalNAc-T1) — curation notes

2026-09-17 — de novo review

No -deep-research-PROVIDER.md: tooling unavailable (OpenAI key rejected);
nothing self-authored was named as provider output. Grounded in the cached
publications plus two papers fetched during review.

Core activity and the paralogue problem

Four independent IDAs support GO:0004653. The specificity study is the one that
matters for how far the term may be pushed: PMID:9295285 and PMID:9295285.

The family-level IBA is well placed (36 WITH/FROM entries spanning fly, worm,
mouse, rat, Xenopus and fourteen human paralogues) but it asserts the shared
activity. What distinguishes the twenty human GALNTs is which peptide sites
each prefers, and GO does not currently express that. This is the axis on which
paralogue over-annotation in this family would occur; raised in
suggested_questions.

Golgi localisation — prefer the GALNT1-specific reference

Originally the Golgi rows leaned on PMID:12506059, an isoform-antibody survey of
ocular surface epithelia (rated relevance: MEDIUM here). UniProt's own
RP SUBCELLULAR LOCATION reference for the entry is better and is now cited:

• PMID:9394011
• PMID:9394011

Three independent methods, GALNT1-specific. The "throughout the Golgi stack"
finding also supports GO:0032580 Golgi cisterna membrane.

Note GO:0000139 Golgi membrane and GO:0032580 Golgi cisterna membrane are
not parent/child: GO:0000139 is not among GO:0032580's ancestors under
is_a or is_a,part_of (checked via QuickGO). Both are kept.

Extracellular region — downgraded, and why

GO:0005576 was initially ACCEPTed on the reasoning that UniProt models the
soluble form as its own curated chain (41..559, PRO_0000012257) rather than as
an alternative location for the intact enzyme, making it a deliberate curatorial
act. That treated an assertion as evidence. Three checks changed the call:

1. The UniProt Secreted line carries no ECO code. (So does the competing
   Golgi location line, so the absence of a code is not by itself the
   discriminator — the difference is that the Golgi location has independent
   experimental support and the secreted one has none.)
2. No GALNT1-specific secretion or extracellular-localisation study was found.
3. The shedding mechanism that would justify the term is characterised, and
   GALNT1 is not among the enzymes it was shown for:
   PMID:35279766.

GALNT1 has zero word-boundary mentions in PMID:35279766 — apparent hits are
substrings of B4GALNT1. Worth recording because the substring trap is easy to
fall into when grepping a family name.

Marked MARK_AS_OVER_ANNOTATED rather than removed: the mechanism is real
(PMID:35279766), so a shed GALNT1 species remains plausible, just unshown.

Viral protein processing

GO:0019082 comes from Reactome's SARS-CoV-2 model (GALNT1 glycosylating ORF3a).
Marked over-annotated: the enzyme's activity does not change with the provenance
of its acceptor, and the same reasoning would give every constitutively expressed
Golgi enzyme a virus-specific annotation. The ERGIC location row from the same
model is kept as non-core, since a location claim is a different kind of thing
from a process claim.

Abstract-only caution exercised

PMID:16638743 is a GalNAc-T3/FGF23 paper cited as an IDA on GALNT1,
abstract-only in the cache. Marked UNVERIFIED rather than MISCITED and the
annotation ACCEPTed: a selectivity claim of that kind is normally established
against a panel of isoforms, so a GalNAc-T1 comparator assay is very likely in
the full text the curator read, and the asserted activity is independently
certain for this gene.