THI22 (Q06490) — Function-Assignment Review: "extracellular region" (GO:0005576)

Gene: THI22 / Saccharomyces cerevisiae (NCBITaxon:559292) / UniProt Q06490
Focus: function_assignment — does THI22 directly have the cellular-component term extracellular region (GO:0005576)?
Seed annotation: GO:0005576, evidence IEA, reference GO_REF:0000044

Executive Judgment

Verdict: Over-annotated / refuted for "extracellular region" specifically — but the protein IS in the endomembrane system (ER + vacuole), not the cytosol.

The GO:0005576 (extracellular region) annotation is a computational carry-over with no experimental support. It originates from a single predicted N‑terminal signal peptide (SignalP-type, ECO:0000255) → UniProt "Secreted" keyword (curator inference, ECO:0000305) → GO:0005576 (IEA, GO_REF:0000044, ECO:0007322). No experiment places THI22 outside the cell.

Pivotal experimental data (Iteration 3): SGD's THI22/YPR121W record states that "SWAT‑GFP and mCherry fusion proteins localize to the endoplasmic reticulum and vacuole respectively." This is genuine (database-level) localization evidence from systematic tagging collections. It shows:
- THI22 does enter the secretory pathway (ER localization observed) — so the predicted N-terminal signal is likely functional, reconciling why the "Secreted" keyword was predicted.
- BUT the observed destinations are ER (GO:0005783) and vacuole (GO:0005773) — endomembrane compartments, not the extracellular space. Entering the ER and accumulating in the vacuole is not secretion out of the cell.
- The competing cytosol (GO:0005829, IBA) annotation is therefore also likely superseded by these imaging data.

Net: "extracellular region" is an over-reach on a genuine secretory-pathway signal; the experimentally supported terms are ER and vacuole. SGD's GO CC set still lists only extracellular (IEA), CC ND (manual), and cytosol (IBA) — the imaging localizations appear in the locus description but are not yet reflected in the GO annotations.

Most important caveats: the ER/vacuole calls come from high-throughput N‑terminal (SWAT‑GFP) and C‑terminal (mCherry) tagging, which can perturb signal sequences and give tag-dependent discrepancies (here ER vs vacuole); primary references should be verified by the curator. I could not run SignalP/DeepLoc programmatically. But no evidence supports the extracellular term.

Evidence Matrix

GO Curation Implications

Lead (requires curator verification): Do not retain GO:0005576 (extracellular region); recommend removal (or NOT-qualification). Replace with the experimentally observed endomembrane compartments.

• The extracellular term is CC, evidence IEA, sourced only from a prediction chain (signal peptide → "Secreted" keyword → GO_REF:0000044 mapping). It is not core and not experimentally supported. Entering the ER is not the same as being secreted to the extracellular region, so the term over-reaches even given a functional signal.
• Preferred CC replacement (experiment-backed): endoplasmic reticulum (GO:0005783) and vacuole (GO:0005773), from SGD's SWAT-GFP and mCherry systematic tagging. Curator should confirm the primary imaging references and appropriate evidence code (HDA/IDA) before annotating.
• Cytosol (GO:0005829, IBA) is likely superseded by the imaging data; treat as lower priority than the ER/vacuole calls.
• If policy retains prediction-derived IEA CC terms, at minimum flag GO:0005576 as low-confidence / prediction-only and note the direct conflict with the ER/vacuole imaging and SGD's ND.
• MF/BP context (not the focus, but relevant for coherence): THI22 sits in thiamine biosynthetic/salvage process (GO:0009228) as a largely redundant, possibly enzymatically inactive paralog; its MF is uncertain ("no HMP‑P kinase activity demonstrated").

This avoids "protein binding" and names specific, experimentally supported compartments.

GO decision table (leads — verify before applying)

Mechanistic Scope

The term under test is a cellular-component (localization) claim. The immediate question is where the THI22 gene product resides. THI22 is a thiaminase‑2/HMP‑kinase family paralog related to the cytosolic thiamine-biosynthesis enzymes THI20/THI21. Unlike them, THI22 carries a functional-looking N-terminal signal and is experimentally observed in the ER and vacuole (endomembrane). The "extracellular" claim is a direct localization assertion generated by sequence prediction, decoupled from any measured secretion. The most that is supported is entry into the secretory pathway (ER); there is no evidence the protein is released to the extracellular space, and no loss-of-function phenotype, interaction, or assay ties THI22 to the extracellular region.

Conflicts and Alternatives

{{figure:nterm_hydropathy.png|caption=N-terminal Kyte–Doolittle hydropathy (window = 9) of THI22 versus paralogs THI20 and THI21. THI22 shows a signal-peptide-like hydrophobic peak (~3.78) confined to a family-atypical ~24-residue N-terminal extension, whereas THI20 (~0.79) and THI21 (~0.49) never reach signal-peptide-like hydrophobicity. This demonstrates the secretion signal is a THI22-specific feature, not a conserved family localization determinant.}}

• Signal is likely functional, but "extracellular" is the wrong endpoint: My N-terminal alignment/hydropathy analysis shows the signal-triggering segment is a ~24-residue hydrophobic extension unique to THI22 (hydropathy peak 3.78 vs ~0.5–0.8 in THI20/THI21); the conserved thiaminase‑2 fold begins only after it. The SGD imaging data (ER localization) indicate this extension does route the protein into the secretory pathway — so it is likely a real ER-targeting signal, not merely a prediction artifact. The error in the seed annotation is the destination: the protein reaches ER/vacuole (endomembrane), not the extracellular space. So GO:0005576 conflates "has a signal / enters secretory pathway" with "is secreted extracellularly."
• Paralog divergence: THI20/THI21 (cytosolic, no signal) diverge from THI22 specifically at this N-terminal extension, so THI22 appears to have neofunctionalized toward endomembrane targeting while its sisters remain cytosolic enzymes.
• Paralog contrast: THI20 and THI21 — same family, cytosolic — lack the signal entirely, indicating the feature is not conserved family biology.
• Database carry-over: UniProt keyword "Secreted" (ECO:0000305 inferred from the prediction) propagated to GO:0005576 by automated mapping; SGD independently declined to annotate extracellular (ND).
• Alternative not excluded: genuine non-classical secretion / cell-wall association (as reported for some "moonlighting" cytosolic yeast metabolic enzymes) — but there is no proteomic secretome/cell-wall evidence cited here for Thi22, so this remains speculative.

Knowledge Gaps

1. Extracellular vs endomembrane. Checked: SGD reports experimental ER (SWAT-GFP) and vacuole (mCherry) localization; QuickGO/UniProt GO CC still lists only extracellular (IEA)/ND/cytosol (IBA). The crux (extracellular?) is now effectively resolved as no — the protein is endomembrane, not extracellular. Remaining gap: reconcile the ER-vs-vacuole tag discrepancy and obtain a GO-annotatable primary reference with an experimental evidence code. Resolve with: curated primary imaging refs, endogenous-tag microscopy, and secretome MS to formally exclude any extracellular pool.
2. Reality of the signal peptide. Checked: UniProt reports ECO:0000255 prediction only. Matters because the whole annotation depends on it. Resolve with: SignalP‑6/DeepLoc/TargetP runs and N‑terminal sequencing / cleavage assay (could not run these tools programmatically here).
3. Whether an upstream start / ORF boundary is correct. Not checked programmatically. Matters because a mis-annotated 5′ extension could spuriously create the hydrophobic N-terminus. Resolve with: Ribo-seq/TSS data and 5′-proteomics.

Discriminating Tests

• GFP/fluorescent-tag microscopy of Thi22 (and side-by-side Thi20) — cytosol vs. secretory/ER/plasma-membrane/extracellular.
• Subcellular fractionation + secretome/cell-wall proteomics MS — is Thi22 peptide detected extracellularly above background?
• SignalP‑6 / DeepLoc‑2 / TargetP‑2 on Q06490 vs THI20/THI21 — quantify signal-peptide probability and cleavage site; compare within family.
• N‑terminal truncation reporter (res 1–19 fused to a secreted reporter) — does the predicted signal actually drive secretion?

Curation Leads (require curator verification)

• Candidate action change: Remove or NOT-qualify GO:0005576 (extracellular region); it is prediction-only (signal peptide → "Secreted" keyword → GO_REF:0000044 IEA) and is contradicted by experimental ER/vacuole imaging. Replace with the experiment-backed ER (GO:0005783) and vacuole (GO:0005773); down-weight cytosol (GO:0005829, IBA).
• Candidate reference/snippet to verify:
• SGD THI22/YPR121W locus summary: "SWAT-GFP and mCherry fusion proteins localize to the endoplasmic reticulum and vacuole respectively." → trace to the primary SWAT‑GFP (N′-tag) and C′-mCherry library papers and confirm evidence code (HDA/IDA) before annotating.
• UniProt Q06490: Signal peptide 1–19 (ECO:0000255); Subcellular location "Secreted" (ECO:0000305, inferred, not experimental); CAUTION "no hydroxymethylpyrimidine phosphate kinase activity could be demonstrated"; FUNCTION "Is not required for thiamine biosynthesis."
• PMID 18028398 (Onozuka et al. 2008): "The C-terminal domains of THI20 family proteins (THI20/21/22) … are homologous to bacterial thiaminase II." — situates the fold in intracellular thiamine metabolism.
• Suggested question for curators: Given experimental ER + vacuole localization, should GO:0005576 be removed and replaced by GO:0005783/GO:0005773? Is the "Secreted" UniProt keyword (and its IEA CC) worth a NOT statement to stop propagation?
• Suggested experiments: endogenous (untagged-locus) fluorescent tagging + colocalization markers to confirm ER/vacuole and resolve the N′-vs-C′ tag discrepancy; secretome/cell-wall MS to formally test for any extracellular pool; signal-peptide cleavage assay to confirm ER entry via residues 1–19.

Provenance: UniProt REST (Q06490, Q08224, Q08975) and EBI QuickGO annotation API were queried live via execute_code; PubMed via search_pubmed (PMID 18028398). Signal-peptide predictors and the yeast GFP-localization database could not be executed programmatically in this environment; those gaps are stated explicitly and not fabricated.

Artifacts

• OpenScientist final report
• OpenScientist final report
• OpenScientist nterm hydropathy