C1GALT1C1 (Cosmc) — curation notes

2026-09-17 — de novo review

No -deep-research-PROVIDER.md exists for this gene: deep-research tooling was
unavailable (the configured OpenAI key is rejected). Per repository policy no
self-authored file was named as provider output. The review rests on the cached
publications, all of which were present for the PMIDs GOA cites, plus five papers
fetched during the work.

The NOT annotation is the interesting row

GOA carries GO:0016263 on Cosmc as NOT|enables. Read without the negation
flag this looks like a textbook necessity-mistaken-for-possession error: the IMP
behind it measured T-synthase activity in patient cells and found it reduced.
It is the opposite — the curators drew exactly the right distinction. Worth
remembering that the review-stub seeding does carry negated: true correctly;
the trap is reviewer-side, in reading the row before the flag.

Cosmc is a pseudoenzyme, not a protein of unrelated fold

An early draft of this review asserted Cosmc has "no glycosyltransferase fold".
That is wrong, and the local UniProt record says so plainly:

• SIMILARITY — "Belongs to the glycosyltransferase 31 family. Beta3-Gal-T
  subfamily." The same family as its client C1GALT1.
• CAUTION — "Was originally (PubMed:12361956) assigned to be a
  glycosyltransferase. However, it was later shown (Ref.2 and PubMed:12464682)
  that it has no transferase activity and rather acts as a chaperone."

So Cosmc is a degenerate GT31 member that retains the family relationship while
having lost catalysis. This matters for the negation's value: family- and
sequence-based inference will keep proposing the activity because it is GT31,
and only an explicit NOT stops it.

Relevant to the project's cazy2go work, where GT31 is already flagged
narrowMatch as too heterogeneous to propagate at family level — a GT31 member
with zero activity is about as strong an argument for that flag as the family
can supply.

ER, not Golgi — and by a mapped signal

Three GOA rows placed Cosmc in the Golgi. The compartment was inherited from its
client's reaction (Reactome models the Golgi-lumenal transfer). Cosmc's own ER
residence is not merely asserted in review prose:

• PMID:21262965
• PMID:21262965
• PMID:18695044

Note UniProt asserts only the generic "Membrane" and commits to neither
compartment, so correcting Golgi → ER does not contradict it.

Chaperone activity is directly demonstrated

• PMID:19923218 — in vitro refolding on purified protein, which isolates
  chaperone activity from any catalytic contribution.
• PMID:21496458

GO:0044183 protein folding chaperone is the term used. Not GO:0051082 unfolded protein binding, which an OpenScientist run recommended: that term is
obsolete (QuickGO returns "obsolete unfolded protein binding"), a known problem
tracked in issue #2222.

Open question carried into the review

Whether a dedicated single-client chaperone should carry involved_in on the
process it enables. GOA itself is unsettled — it uses involved_in on two rows
and acts_upstream_of_or_within on a third. The existing rows are accepted, but
core_functions.directly_involved_in is deliberately not asserted: Cosmc
catalyses no step and contributes neither structure nor cofactor activity to the
glycosyl transfer, which is the same standard used to demote GCNT1's leukocyte
tethering and C1GALT1's angiogenesis in this batch.

Provenance of the two OpenScientist runs

function-hypothesis-go-0016263 (verdict SUPPORTED) and
function-hypothesis-go-0000139 (verdict refuted, i.e. not Golgi). The first
supplied the GT31 correction above; the second surfaced PMID:21262965. Both
reports' claims were re-checked against local records before use, and both
carry reference_review notes recording what each got wrong as well as right.