Working journal for the PAINT-campaign review of human ARHGAP6 (UniProtKB:O43182,
HGNC:676, X p22.3). Append-only.
Deep research: affinage (ARHGAP6-deep-research-affinage.md). Trust gates clear
(accession O43182 matches the local UniProt record, organism human,
self_evaluation_pairwise: win, faith_pct: 100.0).
Affinage's narrative is accurate as far as it goes, and it gets the central point
right: it names the separability of the GAP and actin-remodelling activities and
anchors it to PMID:10699171. But its citation list (11 PMIDs) omits the two papers
that carry the most GO-relevant experimental content, and both are omitted for the
same reason — the title does not say what the paper is about:
| PMID | Why a title-driven search misses it |
|---|---|
| 18434237 | Title misspells the gene as "ARGHAP6" and leads with "phospholipase C-delta1". A search on the correct symbol does not hit the title. |
| 19038263 | Titled for the partner ("novel HERG regulator") and for C. elegans; spells the gene out as "rho-GTPase activating protein 6" rather than ARHGAP6. |
A EuropePMC TITLE:"ARHGAP6" OR TITLE:"ARGHAP6" OR TITLE:"RhoGAPX-1" search returns
17 records and does not include PMID:19038263 — confirming the mechanism rather
than just asserting it. Both papers were found by searching the function
(ARHGAP6 AND phospholipase in PubMed returns exactly these two) and by following
the UniProt DR GO block. Neither paper is in UniProt's RN reference list either:
that list holds the two cloning/function papers (PMID:10699171, PMID:9417914) plus
four large-scale sequence and phosphoproteomics submissions, so it offered no route
to them.
Both are now cached: publications/PMID_18434237.md (abstract only) and
publications/PMID_19038263.md (full text via PMC).
PMID:10699171 (Prakash et al. 2000) is the primary functional paper and is
abstract-only in our cache (full_text_available: false), but the abstract states
the separability directly:
The mouse knockout of the rhoGAP domain alone is normal:
A third output, also GAP-independent, is established by the two missed papers.
ARHGAP6 binds and activates PLCδ1:
and a second, independent lab shows that an ARHGAP6 output runs through PLC and
survives loss of GAP function:
So the picture is one protein with a catalytic RhoA-GAP arm and two arms that do
not need it.
molecular_functionThe GAP-independent actin role is defined entirely by what it is not (not
GAP-dependent) and by cell-biological observation: co-localisation with F-actin
through an N-terminal region, and recruitment of F-actin into growing processes
PMID:10699171. No direct
biochemical activity is demonstrated — no measured actin binding, no nucleation, no
bundling, no capping. Attaching a GAP molecular function would contradict the very
evidence that defines the role; attaching an invented structural or binding MF would
assert a mechanism nobody has measured.
molecular_function is optional on CoreFunction, so the slot is left empty,
and the entry's own description says why. This is a statement about the evidence,
not a claim that GO lacks a suitable term — a negative claim about the ontology is
not needed here and is not made.
See ARHGAP6-bioinformatics/RESULTS.md (check_arginine_finger.py).
TTEGIFRRSANTQ) is identical to the window around UniProt Q07960 position 282.Arginine finger SITE feature agree, and R433 is the residueNOT|enables GO:0005096 IDAenables
GO:0005096 IBA. OCRL substitutes Q757 and carries a positive IDA.What actually supports the GAP call is the functional evidence (the R433G mutant
fails to clear stress fibres), not the residue. And because R433G is a clean
loss-of-GAP reagent, it is also what makes the GAP-independence claims arguments
from a working mutant rather than from a residue being present:
PMID:19038263.
See ARHGAP6-bioinformatics/RESULTS-goa-divergence.md
(check_goa_uniprot_divergence.py).
UniProt's cross-reference block lists 15 GO terms; QuickGO/GOA lists 14; three are in
UniProt only:
GO:0016004 phospholipase activator activity (IDA:BHF-UCL)GO:0043274 phospholipase binding (IPI:BHF-UCL)GO:0141214 positive regulation of phospholipase C/protein kinase C signal transduction (IDA:BHF-UCL)This is not term obsolescence: all three are active on both OLS4 and
api.geneontology.org (checked on two independent services, because these are known
to disagree; QuickGO also reports them active but that would have been a third
endpoint of a service already consulted).
And the missing rows are still load-bearing. GOA is currently serving six ISO
annotations — mouse Arhgap6 (O54834, GO_REF:0000119) and rat Arhgap6 (A0A8I6AAE4,
GO_REF:0000121) — whose WITH/FROM column names O43182 as the source for exactly
these three terms. The projections are orphaned: their stated human source is not in
the human record any more.
Because just fetch-gene seeds existing_annotations from the GOA TSV, these rows
would otherwise never be reviewed at all. They are therefore entered as action: NEW
— not as new curation, but as pre-existing BHF-UCL curation that has fallen out of
the feed the review is built from. The reason field on each says so.
GOA carries enables GO:0005096 TAS from Reactome:R-HSA-9018806 "RAC3 GAPs
stimulate RAC3 GTPase activity". Reactome's own summary places ARHGAP6 in the top
tier — "shown to bind RAC3 and stimulate its GTPase activity" — citing a single
paper, "Li et al. 2016", which resolves to PMID:26628301. That paper's only Rac3
experiment is a co-IP:
It never measures GTPase activity. Reactome's other ARHGAP6 event,
R-HSA-8981637 ("RHOA GAPs"), is by contrast backed by Prakash et al. 2000
(PMID:10699171) and Müller et al. 2020 (PMID:32203420), a family-wide
substrate-specificity screen PMID:32203420.
Action taken: ACCEPT, not REMOVE or MARK_AS_OVER_ANNOTATED. The GO term
GO:0005096 is substrate-agnostic and ARHGAP6 genuinely enables it (for RhoA), so
the row's assertion is true; what is unsupported is the route Reactome took to it.
Unsupported is not contradicted. The problem is recorded in review.reason and in
propagation_review (root_cause: SOURCE_WEAK_OR_INFERRED,
failure_modes: [SOURCE_EVIDENCE_WEAK]) rather than by downgrading a true statement.
GO cannot express the distinction directly. GO:0005096 has zero descendants —
confirmed on two independent services (OLS4 hierarchicalChildren = 0;
api.geneontology.org subgraph descendants = 0). QuickGO returns one child,
GO:1902773 GTPase activator complex, but via capable_of, and it is a cellular
component, not a specialisation of the activity. So there is no "Rho GTPase activator
activity" term to MODIFY to. The GO mechanism for substrate lives on the enzyme as
has input; raised as a suggested_questions entry.
See ARHGAP6-bioinformatics/RESULTS-pdz-interactome.md
(check_pdz_interactome.py). 22 of the 44 GOA rows are GO:0005515 protein binding
IPI from one reference, PMID:36115835 (Gogl et al., a quantitative fragmentomics /
holdup assay). All 22 of 22 partners carry PDZ domains, and ARHGAP6 ends in
...LPETLV, a canonical class I PDZ-binding motif (-X-S/T-X-Φ: T at −2, V at 0).
So these are one binding determinant measured once, not 22 independent findings.
Action: all 22 → MODIFY, replacement GO:0030165 PDZ domain binding. The
repository policy (.claude/skills/annotation-reviewer/SKILL.md, Quality Standards,
added 2026-09-12 in b1cff30014) is that GO:0005515 must not simply be kept:
MODIFY when the cited paper supports a more informative molecular function, and
otherwise REMOVE. An earlier version of this review used KEEP_AS_NON_CORE, which
the policy does not offer; that is corrected here.
MODIFY rather than REMOVE because the cited paper genuinely does support a more
informative term. See RESULTS-binding-evidence.md (check_binding_evidence.py):
IntAct records all 22 pairs as direct interaction, detected by holdup assay
(MI:2437) — and for MAST2 and SNX27 additionally by fps (MI:0053, fluorescence
polarization spectroscopy). The assay is a recombinant PDZome library against a
library of C-terminal PDZ-binding motif peptides, so what was measured is binding
between ARHGAP6's C-terminal motif and a PDZ domain. GO:0030165 restates that; it
does not infer a function from the interaction.
Nothing beyond the binding is asserted — no adaptor, scaffolding, localisation or
signalling role — because interaction evidence alone would not support one, and the
policy says so explicitly. Worth recording: the cached full text of PMID:36115835
never names ARHGAP6 (the pairs are in the supplementary peptide library), so the
per-pair evidence comes from IntAct's curated record of that paper, not its prose.
The HERG/IKr result (PMID:19038263). Well controlled — overexpression plus shRNA
knockdown of endogenous ARHGAP6 in an atrial myocyte line, with cell-surface
quantification — and mechanistically the most interesting GAP-independent output.
It is not proposed as a NEW annotation: it is one paper from one lab, largely in
heterologous systems, and the effect is routed through PLC rather than exerted
directly on the channel, which makes it an indirect effect of the kind the repository
guidance asks us not to annotate. It is recorded as four findings entries on
references[PMID:19038263](https://pubmed.ncbi.nlm.nih.gov/19038263) (not merely asserted in prose — the entries are in the
YAML), described in the PLC core function, and raised as a suggested_questions /
suggested_experiments item instead.
The cancer-phenotype literature. Four further papers were cached while working on
this gene and none yields an annotation. Each now carries a reference_review saying
so, rather than sitting in publications/ unexplained:
| PMID | Why no annotation |
|---|---|
| 30816546 | Lung-cancer cell-line migration/invasion and phospho-STAT3, under manipulated expression. Several steps downstream of the GAP activity. |
| 33116826 | Apoptosis, glycolysis and cisplatin sensitivity attributed to STAT3. Same objection. |
| 38287795 | Ferroptosis via RhoA–ROCK1–p38 MAPK. Consistent with ARHGAP6 being a RhoA GAP, which is worth noting, but three signalling steps removed and measured only under manipulated expression. |
| 12673365 | A methods paper (E. coli expression and His-tag purification). Makes no functional claim at all. |
These are the bulk of affinage's dated-findings table, so the decision to exclude them
is where most of affinage's content ends up — recorded rather than silently dropped.
All five analyses live in ARHGAP6-bioinformatics/ and each takes --self-test,
which mutation-tests its guards (every guard must fire with its expected message,
every negative control must stay silent):
cd genes/human/ARHGAP6/ARHGAP6-bioinformatics
uv run check_arginine_finger.py --self-test && uv run check_arginine_finger.py
uv run check_goa_uniprot_divergence.py --self-test && uv run check_goa_uniprot_divergence.py
uv run check_pdz_interactome.py --self-test && uv run check_pdz_interactome.py
uv run check_binding_evidence.py --self-test && uv run check_binding_evidence.py
uv run check_review_consistency.py --self-test && uv run check_review_consistency.py
The first four regenerate their RESULTS*.md. The last regenerates nothing: it
re-tests this review's own claims against the files — in particular that every
file: supporting_text occurs verbatim and exactly once in the file it names,
which the repository's reference validator does not check (it validates quotes
only for PMID:/DOI: references). It also re-tests the counted claims made in
this notes file and in the PR, so a sentence here cannot quietly stop being true.
Expected output today:
consistent: 49 file: quotes verbatim and unambiguous, 25 references adjudicated, counted claims hold
GO:0030041 actin filament polymerization (NAS, PMID:10699171) →MARK_AS_OVER_ANNOTATED. The paper shows recruitment of existing F-actin intoGO:1902533 positive regulation of intracellular signal transduction (IEA,enables GO:0005096 IDA on mouse Arhgap6 cites PMID:25211221, a paper titled