PDI1 review notes

Update 2026-09-02

Audited PDI1-ai-review.yaml for factual accuracy. The top-level description
field asserted "Orthologous to human PDIA3." This is incorrect and has been
corrected: the closest human counterpart of yeast Pdi1p is PDIA1/P4HB.

Evidence that does not discriminate (recorded so it is not reused as an
argument):

• Recommended name and EC are shared across the whole catalytic PDI family and
  cannot separate P4HB from PDIA3. Yeast PDI1 (PDI1-uniprot.txt) is
  DE RecName: Full=Protein disulfide-isomerase; / DE EC=5.3.4.1;,
  human P4HB (P07237) is the same, and human PDIA3 (P30101) also carries
  EC=5.3.4.1 {ECO:0000269|PubMed:27897272, ECO:0000269|PubMed:7487104}
  (checked against the UniProt REST record for P30101 on 2026-09-04). An
  earlier draft of these notes claimed PDIA3 had "no independent EC entry of
  its own in the same catalytic sense"; that was wrong and has been deleted.
• eggNOG KOGs are family-level groups. Yeast PDI1 and human P4HB both
  cross-reference eggNOG; KOG0190; Eukaryota., but so does human PDIA3
  (P30101). Shared KOG0190 membership is therefore consistent with the
  assignment but is not confirmation of orthology, and the earlier
  "confirming direct orthology" wording has been removed.
• Shared a-b-b'-a' domain architecture. Yeast Pdi1p has
  FT DOMAIN 29..141 "Thioredoxin 1" and
  FT DOMAIN 356..485 "Thioredoxin 2"; human P4HB has
  FT DOMAIN 18..134 "Thioredoxin 1" and
  FT DOMAIN 349..475 "Thioredoxin 2" (FT lines verbatim from the
  respective -uniprot.txt files). An earlier draft listed this as a
  discriminating argument; that was wrong, because the same layout is shared
  with PDIA3 PMID:17507649 — ERp57 is PDIA3. What the matched FT lines do establish is that
  Pdi1p is a canonical two-active-site PDI, which rules out the b-type-only
  and single-a-domain family members (ERp27, ERp29 and similar) but says
  nothing about P4HB versus PDIA3.
• Partnership in the Ero1-driven oxidative-folding relay. Pdi1p activates
  Ero1p, which reoxidizes it, and human P4HB likewise
  Interacts with ERO1B (PubMed:11707400) (P4HB-uniprot.txt:418). An earlier
  draft listed this as a discriminating argument, with the PDIA3 half marked
  "not checked" because P30101 is uncached. The interaction is symmetric,
  however, and both human Ero1 paralogs are cached here — both list PDIA3,
  with the P4HB and PDIA3 entries adjacent and at comparable evidence depth:
  ERO1A-uniprot.txt:352-353 reads
  Q96HE7; P07237: P4HB; NbExp=3; IntAct=EBI-2564539, EBI-395883; then
  Q96HE7; P30101: PDIA3; NbExp=3; IntAct=EBI-2564539, EBI-979862;, and
  ERO1B-uniprot.txt:198 reads
  Q86YB8; P30101: PDIA3; NbExp=2; IntAct=EBI-2806988, EBI-979862; — ERO1B
  being the exact partner cited for P4HB. Ero1 partnership therefore does not
  separate P4HB from PDIA3 on any data cached here. Two caveats on how far this
  goes: an IntAct binary interaction is not a statement about preferential
  relay flux (a claim that an Ero1 paralog prefers PDI over ERp57 would need its
  own citation and is not made here), and demoting this bullet does not touch
  the conclusion — the substrate-scope contrast below remains decisive.

Evidence that does discriminate:

• General versus glycan-restricted substrate scope. Yeast Pdi1p is
  "required for formation of disulfide bonds in secretory and cell-surface
  proteins and which unscrambles non-native disulfide bonds", partnering EPS1,
  KAR2 and MNL1 (PDI1-uniprot.txt). PDIA3 is instead specialized on
  N-glycosylated clients through the lectin chaperones — it
  Interacts with ERP27 and CANX and is a "Core component of the major
  histocompatibility complex class I (MHC I) peptide loading complex"
  with CALR (P30101; these P30101 lines are quoted from the same UniProt REST
  record fetched on 2026-09-04 that the EC check above used). Yeast has no such
  calnexin/calreticulin-partnered PDI specialization for Pdi1p. This split is independently corroborated in the
  literature PMID:17507649, against PDI itself catalyzing "in vitro redox reactions in a wide
  variety of substrates". This is the decisive argument in this audit.
• Caveat on framing: the PDI family expanded independently in the fungal and
  metazoan lineages, so this is best stated as "closest human counterpart"
  rather than a clean 1:1 orthology relationship.
• The gene's own PDI1-deep-research-falcon.md file frames the mammalian
  comparison around PDIA1, not PDIA3: "Although much of this work is framed
  around mammalian PDIA1, the mechanistic concepts apply to canonical yeast
  PDI architecture and redox cycling."

Repo-local data pointing the other way (recorded so this audit is not silent
about the datapoints that disagree with its conclusion):

• PANTHER places yeast PDI1 in the same subfamily as PDIA3, not as P4HB.
  PDI1-uniprot.txt:238 reads
  DR PANTHER; PTHR18929:SF132; PROTEIN DISULFIDE-ISOMERASE A3; 1., and
  interpro/panther/PTHR18929/PTHR18929-entries.csv assigns P30101 (PDIA3) to
  PTHR18929:SF132 as well, whereas P07237 (P4HB) sits in PTHR18929:SF101
  (P4HB-uniprot.txt:716). This is the strongest contrary datapoint, and it is
  on the same UniProt file quoted elsewhere in these notes.
• OrthoDB splits Pdi1p from P4HB. PDI1-uniprot.txt:206 is
  DR OrthoDB; 427280at2759; -. while P4HB-uniprot.txt:627 is
  DR OrthoDB; 72053at2759; -. — different eukaryotic orthologous groups.

Neither datapoint overturns the "closest human counterpart is PDIA1/P4HB"
conclusion, but the reasons should be stated rather than assumed. The PANTHER
call is an automated HMM subfamily assignment, not a curated orthology
statement; given the lineage-independent expansion of the PDI family noted
above, there is no metazoan subfamily that the single fungal canonical PDI fits
cleanly, so a best-scoring-model assignment is weak evidence of orthology
either way. The functional argument that runs the other way is the decisive
one, and it is the only one the conclusion rests on: general versus
glycan/lectin-restricted substrate scope. The description wording ("closest human
counterpart", not "ortholog of") is deliberately chosen to remain compatible
with this unresolved family-level ambiguity.

Possibly discriminating, not yet checked: Pdi1p and P4HB share an identical
OMA group fingerprint — DR OMA; FFGMKKD; -. at PDI1-uniprot.txt:205 and
P4HB-uniprot.txt:626. Unlike the family-level KOG0190 above, an OMA group is
a real orthologous-group call, so this would be genuine evidence if P30101's
fingerprint differs. P30101 is not cached in this repo (there is no
genes/human/PDIA3/), so that comparison has not been run; recorded as a lead
only, not as evidence.

No GO term annotations were affected by this fix (the existing_annotations
list does not reference cross-species orthology); only the standalone
biological description field was corrected. Per CLAUDE.md the description
is a standalone biological summary, so the supporting identifiers (eggNOG KOG,
EC numbers) and the explicit "not PDIA3" negation are kept here rather than in
that field.