GCNT1 (C2GnT1) — curation notes

2026-09-17 — de novo review

No -deep-research-PROVIDER.md: deep-research tooling was unavailable (OpenAI
key rejected). Per repository policy nothing self-authored was named as provider
output. Review grounded in the cached publications, which existed for every PMID
GOA cites.

Core activity is unusually cleanly established

PMID:1329093 — and, importantly
for a large family where paralog mis-assignment is the standing risk,
PMID:1329093. The negative specificity control is what
makes the specific MF term safe here.

Nuclear speckles: an antibody artefact, and how it was established

GO:0016607 nuclear speck came from GO_REF:0000052 (immunofluorescence
curation). Removed. The topological argument is a priori — the catalytic domain
occupies residues 33–428 on the lumenal face, leaving a nine-residue cytoplasmic
tail, so a membraneless nuclear body is not reachable — but the decisive evidence
is empirical, from querying the Human Protein Atlas API directly:

(B3GNT6, ST3GAL1, GALNT1, MGAT1 have no informative HPA IF record.)

The distinction that matters is replace vs supplement. GCNT4, ST6GALNAC1 and
FUT8 each carry a nucleoplasm background call alongside a retained Golgi call —
ordinary nuclear background. GCNT1's nuclear call replaces the Golgi
entirely, i.e. the antibody misses the compartment where this enzyme's
biochemistry demonstrably occurs. That is the signature of off-target binding.

Correction to note: an OpenScientist run claimed GCNT1 uniquely lacks a
Golgi call and listed C1GALT1 and GCNT2 as having no HPA data. Both have records,
and C1GALT1 also lacks a Golgi call. Two related core-1/core-2 enzymes showing
the same pattern strengthens the artefact reading — it looks like an
antibody-class problem — but only GCNT1's version was imported into GO. Worth a
broader audit; raised in suggested_questions.

Guilt-by-substrate, concentrated

GCNT1 carries leukocyte tethering or rolling, its positive regulation, and cell
adhesion molecule production, all because it builds the core 2 branch of a
selectin ligand. Two things from reading the full text of PMID:23027862:

• The tethering/rolling/ICAM-1 assays were run on GOLPH3-depleted cells, not
  GCNT1-depleted cells — a two-step inference through enzyme mislocalisation.
• Direct GCNT1 knockdown was done, for the biosynthetic readout only:
  PMID:23027862, with PSGL-1 protein level unchanged.

So the biosynthetic process annotation is directly supported; the cell-behaviour
terms are downstream and non-core. GO:0060352 cell adhesion molecule production
is the weakest: the paper's own control shows the adhesion molecule is still
produced, only differently glycosylated.

Gated lead — possible shedding

PMID:35279766 — GCNT1 peptides appear among Golgi enzymes whose abundance
tracks SPPL3 expression. GCNT1 has no extracellular annotation in GOA. Recorded
as a suggested_questions lead only, not asserted: a single peptide
observation in a study focused on SPPL3 is not sufficient, and the same standard
was applied in marking GALNT1's extracellular annotation over-annotated.

Localization nuance

GO:0005802 trans-Golgi network marked over-annotated: the paper reports
PMID:23027862 and PMID:23027862, i.e. distribution across the
stack with a cis-lean, which GO:0031985 Golgi cisterna describes better.