STING1 (TMEM173) review journal

Reviewer: AI review, 2026-09-17. UniProt Q86WV6, HGNC:27962 (STING1; TMEM173 retired).
250 GOA rows / 64 distinct GO terms. Strategy: one decision per GO term, applied to all rows
of that term (project rule: same term -> same action regardless of evidence code).

1. Settled baseline (uncontested)

STING1 is a four-transmembrane ER membrane protein with a cytosolic ligand-binding /
signalling domain. It is the receptor for cGAS-derived 2'3'-cGAMP and for bacterial cyclic
dinucleotides; ligand binding drives high-order oligomerisation, COPII-dependent ER exit to the
ERGIC/Golgi, TBK1 recruitment and phosphorylation of the STING C-terminal tail, IRF3 docking,
and type I IFN plus NF-kB transcriptional output.

2. The contested/new question: is there a genuine phosphoinositide-binding MF?

Answer: yes, and GOA already carries it. GO:0080025 phosphatidylinositol-3,5-bisphosphate binding is already present as IDA from both 2026 Nature papers, so this is not a term I had to
introduce. What I added as NEW is GO:0005546 phosphatidylinositol-4,5-bisphosphate binding,
which the companion structural paper establishes with the same rigour but which GOA has not
captured.

Evidence, paper by paper:

How much weight should companion papers from one group carry?

These are co-submitted companion papers with overlapping authorship (Tan JX, Chen ZJ and Bai XC
appear on both; the cryo-EM structure quoted in the Tan paper is the Li paper). Tan et al. say so
explicitly: PMID:41639454
So the two papers are corroboration within one research programme, not independent replication.
No second laboratory has yet reproduced the direct lipid binding.

Three considerations nevertheless argue for accepting the MF rather than holding it as UNDECIDED:

  1. The evidence is multi-modal within the programme, not a single assay. Direct binding (FRET;
    lipid strips), structure (cryo-EM density at a defined dimer-dimer groove), structure-guided
    loss-of-function (K20E/R71H abolish PIP binding; S80I reduces cholesterol binding), in vitro
    liposome reconstitution, and cell-based phenotypes (trafficking and pS366/pTBK1/pIRF3 loss,
    PIKFYVE deletion phenocopy). A single artefact does not explain all of these.
  2. It converges on pre-existing, independent human genetics. R71 is part of the common HAQ
    allele: PMID:41639452 The hypofunction of HAQ was known long before and from
    other groups; the lipid site explains it.
  3. A lipid role in STING activation was already reported independently. The authors frame their
    work as mechanism for prior observations: PMID:41639452 and PMID:41639452

Two independent commentaries treat the finding as established rather than contested:
PMID:41997109 and PMID:41997116 Commentaries are not replication either, but
they establish that the field has not raised a methodological objection.

Caveat I am recording rather than resolving: PtdIns(4,5)P2 binds the same groove with similar
potency in vitro, yet it is a plasma-membrane-enriched lipid and activated STING does not traffic
to the plasma membrane. The authors themselves flag this: PMID:41639452 So GO:0005546 is a well-supported
binding capability whose physiological occupancy is not established; I annotate the binding and
say so in the reason rather than inflating it into a process claim.

Term choice. I used the two specific terms that match the two lipids actually assayed
(GO:0080025, GO:0005546) rather than the parent GO:1902936 phosphatidylinositol bisphosphate binding, because both specific species were tested individually and separately structurally
resolved. GO:1902936 remains the right fallback if a curator prefers one annotation over two.
PtdIns4P was explicitly the weak/likely-indirect case and I did NOT annotate it:
PMID:41639452

Publisher Correction PMID:41721038 — what I could and could not establish

PMID:41721038 is Publisher Correction: PtdIns(3,5)P(2) is an endogenous ligand of STING in innate immune signalling, Nature 2026 Mar;651(8105):E11, doi 10.1038/s41586-026-10280-6, erratum for
PMID:41639454. I could not retrieve what it corrected. PubMed carries only the erratum notice
with no descriptive text; Europe PMC returns no abstract for it; the Nature page is behind an
authentication redirect and the aggregator copies reproduce only the title. The only substantive
thing I can say without guessing is categorical: Springer Nature reserves "Publisher Correction"
for errors introduced in production by the publisher (as opposed to "Author Correction" for
author-introduced errors and "Retraction" for invalidating problems), so on its face it is not a
correction to the data or conclusions. I have deliberately NOT inferred what was changed, and the
annotations do not lean on the correction either way.

3. Proton channel activity (GO:0015252) — scrutinised, and it holds

This is a genuinely non-obvious assignment for an ER adaptor, so I checked what it rests on. It is
not a single-lab claim, unlike the lipid finding:

I searched PubMed for a refutation and found none. Four labs and a medicinal-chemistry programme
converge; GO:0015252 is accepted and treated as a second core molecular function, distinct from
the adaptor activity and responsible for a separable output branch (LC3B/GABARAP lipidation,
inflammasome, TFEB-driven lysosome biogenesis). Note the contrast with the other gene in this
batch, TMEM175, where the proton-channel claim is genuinely contested; for STING1 it is not.

4. Transcription coactivator activity (GO:0003713) — the one MF I demoted

GO:0003713 transcription coactivator activity (IDA, PMID:18818105, Zhong et al. 2008 Immunity,
the MITA paper) is wrong for a four-pass ER membrane protein. GO defines a transcription
coactivator as a coregulator that acts on a DNA-bound transcription factor at the promoter; STING
never enters the nucleus. The underlying observation is real and is an IDA I have no grounds to
delete — MITA overexpression activates IFN-beta reporters and MITA binds IRF3
PMID:18818105 — but that is adaptor/scaffold behaviour. Action: MODIFY to GO:0035591 signaling adaptor activity, which GOA already carries from six other papers. Same reasoning demotes
GO:0045944 positive regulation of transcription by RNA polymerase II (same paper) to
GO:0032481 positive regulation of type I interferon production.

The related GO:0061629 RNA polymerase II-specific DNA-binding transcription factor binding (IPI,
same paper) is not wrong — IRF3 is exactly such a factor — so it is kept, as non-core.

5. Mitochondrial outer membrane (GO:0005741) — over-annotation, not removal

Six rows, four of them EXP (PMID:18724357, PMID:19285439, PMID:19433799, PMID:19776740) and one IDA
(PMID:18818105); UniProt still lists "Mitochondrion outer membrane" with four ECO:0000269 refs. All
date from 2008-2009. The modern consensus, including the same UniProt entry's own FUNCTION block
and every structural/trafficking paper since, is that STING is an ER membrane protein that exits to
ERGIC/Golgi. Project rules forbid REMOVE on experimental annotations whose full text I have not
read, and mitochondria-associated ER membrane (MAM) contact sites plausibly explain a genuine
mitochondrial-fraction signal in the original work. Action: MARK_AS_OVER_ANNOTATED on all six.

6. Other locations

7. IBA handling

Eleven terms carry an IBA row (GO:0000045, GO:0002218, GO:0005776, GO:0005789, GO:0016239,
GO:0032481, GO:0035438, GO:0045087, GO:0051607, GO:0061507, GO:0061709). All are biologically
defensible for human STING1 and I did not inspect the PANTHER tree (PTN005046674) beyond the
WITH/FROM lists, so all are ACCEPTed rather than downgraded — downgrading would require
propagation_review metadata I would have to invent. The one I am least comfortable with is
GO:0061709 reticulophagy; UniProt supports it only "By similarity" (c-di-GMP-triggered
reticulophagy), so the reason records that it is the weakest member of the autophagy cluster.

8. Review-based NAS annotations (PMID:40861013)

Four terms come as NAS from one 2025 review, "Beyond interferons: Non-canonical roles of
MITA/STING" PMID:40861013. GO:0006914 autophagy is
redundant with the better-supported GO:0016239 and was modified to it; GO:0090398 cellular senescence is well grounded elsewhere in the literature and is kept as non-core;
GO:0045820 negative regulation of glycolysis and GO:0055088 lipid homeostasis are downstream
metabolic consequences asserted only in review prose and are marked as over-annotation.

9. Validation

uv run --no-dev ai-gene-review validate --verbose --terms genes/human/STING1/STING1-ai-review.yaml
prints ✓ Valid (no warnings). 251 annotations over 65 terms: 165 ACCEPT, 40
MARK_AS_OVER_ANNOTATED, 31 MODIFY, 13 KEEP_AS_NON_CORE, 1 REMOVE (GO:0005777 peroxisome),
1 NEW (GO:0005546 PtdIns(4,5)P2 binding). No GO term carries two different actions.

10. Core vs non-core, as decided

Core: cGAMP/CDN binding (GO:0061507, GO:0035438) -> signaling adaptor activity (GO:0035591)
at ER/ERGIC/Golgi membranes, driving GO:0140896, GO:0032481 and GO:0043123; and proton channel
activity (GO:0015252) driving the non-canonical autophagy/lysosome branch. Lipid binding
(GO:0080025, GO:0005546, GO:0015485) is recorded as a core-adjacent activation requirement — it is
a genuine MF of the protein but it gates the adaptor function rather than constituting a third
output.

Non-core: senescence, TORC1 regulation, lysosome organization, cytokine production in general,
dsRNA responses, ubiquitin-ligase binding, IRF3 (TF) binding, self-oligomerization as a BP.