VPS74 (Q06385, YDR372C) — curation notes

Provenance note: no deep-research provider API key was available in this
container, so just deep-research could not be run and there is no
VPS74-deep-research-*.md file. The literature below was read directly from the
cached publications/PMID_*.md records (most have full_text_available: true),
plus the UniProt flat file and SGD locus records. The GONUTS/CACAO wiki
(gowiki.tamu.edu) was queried on 2026-10-29 for PMID:24019977 and
PMID:18410729 and redirects to maintenance.html ("temporarily offline"),
matching the status recorded in projects/CACAO.md; both CACAO papers are
full-text cached here, so the wiki was not needed.

Identity

Core mechanism: a cytosolic sorting receptor for Golgi glycosyltransferases

Schmitz et al. 2008 established the mechanism. Vps74p binds the short N-terminal
cytosolic tails of cis/medial Golgi mannosyltransferases
PMID:18410729,
assayed as GST fusions carrying the first 15 residues of KRE2, MNN2, MNN5 and
OCH1 (Fig. 3). Loss of VPS74 mislocalizes those enzymes to the vacuole
PMID:18410729,
while multi-TM Golgi proteins, TGN proteins and the ER glycosyltransferase Alg5p
are unaffected. The protein tetramerizes through a β-hairpin (197–208); deleting
it abolishes both tetramer formation and function.

Tu et al. 2008 independently defined a pentameric tail motif and showed Vps74p
also binds COPI
PMID:18635803,
proposing retention by promoting incorporation into COPI vesicles
PMID:18635803.
The coatomer-binding determinant is an N-terminal arginine cluster
PMID:22889169,
and losing it inactivates retention without much affecting membrane binding
PMID:22889169.

So the molecular function is a cargo adaptor / sorting receptor (GO:0140312):
one face reads glycosyltransferase cytosolic tails, another reads COPI, and the
GPP34 fold reads PtdIns4P.

PtdIns4P binding and Golgi recruitment

Vps74p is a PI4K (Pik1) effector. Direct lipid binding was shown by dot blot
PMID:20026658
and independently by Dippold et al.
PMID:19837035.
Golgi association requires ongoing PtdIns4P synthesis — in a pik1-83ts strain
PMID:19837035
— and mutating the basic pocket (W88/R97, K178/R181) abolishes both binding and
function in vivo
PMID:20026658.

Second, lipid-regulatory function: Vps74–Sac1

Wood et al. 2012 showed Vps74 binds the Sac1 phosphatase directly
PMID:22553352
and that both are needed to keep medial-Golgi PtdIns4P low
PMID:22553352.
This is the evidence behind GO:0060304 (regulation of phosphatidylinositol
dephosphorylation, IMP with SAC1).

Where in the Golgi?

Quantitatively, the steady-state pool is cis/medial, the opposite polarity to
PtdIns4P itself
PMID:22553352,
consistent with Schmitz et al.
PMID:18410729.
There is also a substantial soluble pool
PMID:18410729.
TGN co-localization is nonetheless real but partial
PMID:19837035
PMID:24019977.

Redundancy and partner pathways

Wang et al. 2020 found a parallel Bre5p/Ubp3p deubiquitinase retention pathway;
Ktr3p and Mnn4p need both, and the Vps74p requirement is nutrient-dependent
PMID:32673164
PMID:32673164.

CACAO rows (ASSIGNED BY = CACAO in VPS74-goa.tsv)

Four rows, two papers.

  1. GO:0005794 Golgi apparatus, IDA, PMID:24019977. Figure 2 images
    GFP-Vps74p against Arf1p-mRFP and Arl1p-mRFP. Term matches the assay
    exactly. → ACCEPT.
  2. GO:0005802 trans-Golgi network, IDA, PMID:19837035. Figure S2B,
    EGFP-Vps74p vs Sec7p-dsRed in yeast. Supported, but the quantitative
    cisternal profiling in PMID:22553352 puts the steady-state peak on cis/medial,
    so the TGN is a real but peripheral location. → KEEP_AS_NON_CORE.
  3. GO:0006487 protein N-linked glycosylation, IMP, PMID:18410729. Figs 1A–1C
    (pulse-chase of secreted proteins, invertase mobility ± PNGase F, mCPY).
    Real data, but the paper itself says the role is indirect
    PMID:18410729.
    Classic phenotype-read-as-process: the mechanism is GO:0045053, annotated by
    SGD from the same paper. → MARK_AS_OVER_ANNOTATED.
  4. GO:0035269 protein O-linked glycosylation via mannose, IMP, PMID:18410729.
    Fig. 1D, Hsp150p-HA mobility. Same shape as (3). → MARK_AS_OVER_ANNOTATED.

So two of four CACAO rows are sound localizations read off real images, and two
are the same downstream phenotype term pair. No wrong-gene, wrong-organism or
sign-inversion problems here; the student picked the right papers and the right
evidence codes.

Other rows worth flagging