A bulk MudPIT inventory of human parotid-gland exosomes. ARHGAP23 is one identification in a large catalogue; the study reports no enrichment, no functional follow-up and no secretion mechanism for it.

A set-level reaction whose catalyst is the defined entity "GTPase activator activity of RHOA GAPs", so every member receives GO:0005096 and GO:0005829 by TAS. Unlike the analogous case for ARHGAP11B in this batch, ARHGAP23's membership is explicitly sourced in Reactome's own summary text, which lists it with a citation.

A pathway-membership projection: every participant in the RHO GTPase cycle receives GO:0051056 by TAS. It says that ARHGAP23 belongs to the cycle, not what it does in it.

The RAC1 counterpart of R-HSA-8981637, with the same set-level catalyst construction. Reactome's summary sources ARHGAP23's membership to two publications, the earlier of which is the Rac FRET biosensor study PMID:28114311 (Martin-Vilchez et al. 2017).

Establishes ARHGAP23 as a RhoA suppressor held at the non-protrusive lateral cortex by the planar cell polarity protein Prickle1. Knockdown of Prickle1 or of ARHGAP21/23 raises GTP-bound RhoA; overexpressing ARHGAP23 reverses the elevation in Prickle1-knockdown cells, to the same degree as the RhoA inhibitor C3 transferase.

Also the clearest evidence against a simple Rac1-GAP reading: in the same experiment, and the same cells, ARHGAP21/23 knockdown decreased rather than increased active Rac1.

Direct cellular evidence for Rac1-directed GAP activity, by Rac FRET biosensor: shRNA against ARHGAP23 raised Rac1 activity to the level of a constitutively active Rac probe. The same study shows ARHGAP23 knockdown yields nascent adhesions and its overexpression increased adhesion maturation, and in rat hippocampal neurons two independent shRNAs produced immature, filopodial spines.

Overexpression phenotype at focal adhesions, the reciprocal of the knockdown result.

A cellular FRET-biosensor screen over the human RhoGEF/RhoGAP family. In the Fig. 1b Source Data, ARHGAP23 scores significantly on RhoA (normalised delta-R/R0 -0.364, p = 2.9e-05) and on Rac1 (-0.372, p = 2.9e-06) and not on Cdc42 (+0.197, no p-value, flag 0), and Supplementary Table 2 records the resulting call as the dual RhoA+Rac1 class. The same table's literature columns list nothing under "in vitro". These numbers are parsed from the publisher's open Source Data by the committed script, not transcribed.

Its Supplementary Table 4 describes the ARHGAP23 TIRF construct as a GAP-deficient arginine finger mutant, and the supplementary figure names it ARHGAP23-R986K. Position 986 is a real arginine in the canonical sequence, and the screened cDNA (NP_001186346.1, 1491 aa) is the same length as Q9P227, so the numbering is comparable - but 986 is not the arginine finger. It aligns to ARHGAP1 R323, which sits in the 1TX4 interface but does not contact the transition-state ligands, whereas the annotated finger ARHGAP1 R282 (equivalent to ARHGAP23 R942) does. The mutant was used for localization only; the proteins whose arginine-finger mutants validated the activity screen were ARHGAP11A, ARHGAP40, ARHGAP4, FAM13A and SYDE2.

ARHGAP23 co-purifies with plakophilin 4 from keratinocytes while its close paralog ARHGAP21 does not, which is the cleanest published discrimination between the two proteins.

Functional consequence: in PKP4-knockout keratinocytes ARHGAP23 depletion abolishes stress fibres, placing an active cytoplasmic ARHGAP23 pool upstream of cortical RhoA suppression.

Localization: overexpressed ARHGAP23 is found at the cell cortex, independently of the Prickle1-dependent lateral cortex result in PMID:27226243.

ARHGAP23 is one of four proteins identified as core VE-cadherin interactors in endothelial cells. The paper's co-immunoprecipitation and co-localization follow-up was performed on ARVCF, not on ARHGAP23, so this is an interaction result and not a localization result for this gene.

The original in-silico identification of the gene and its PDZ-PH-RhoGAP domain architecture, and the source of UniProt's tissue-specificity statement. It contains no functional assay.

The experimental grounding of the PANTHER node from which ARHGAP23's IBA descends: this is the paper behind ARHGAP21's own enables GO:0005096 IDA. Its measured substrate is Cdc42 at the Golgi, which is not the substrate profile later measured for ARHGAP23.

The systematic in-vitro kinetic study of the human RHOGAP family. It is cited here for a negative: ARHGAP23 is not among the representatives it purified and assayed, which is why this review states that no purified-protein GAP assay exists for this gene.

ARHGAP23's PDZ domain could not be amplified and is absent from the human PDZome resource, so no experimental PDZ-ligand profile exists for this protein.

ARHGAP23 is the named gene at the 17q12 eosinophilic-esophagitis risk locus. A genotype-phenotype association with no functional follow-up on the protein.

The only in-vivo genetic handle on the gene: Arhgap23 transcript is reduced in lymphatic-endothelial Foxc1;Foxc2 double-knockout mouse embryos, and FOXC1/FOXC2 bind conserved regions of the locus. This is upstream transcriptional regulation of the gene, not evidence about the protein's activity.

A database-mining expression analysis. It performs no experiment on the protein, and its opening functional premise inverts the direction of GAP action.

The machine-fetched Affinage record. Its trust gates passed (accession Q9P227, human, self-evaluation "win"), and its mechanistic narrative is accurate: it identifies ARHGAP23 as a cytoplasmic RhoA suppressor regulated by PKP4 at junctions, and its own grounding lists cytosol and plasma membrane as the localizations.

Its coverage, however, is three citations, and its closing claim is false. Three papers that directly measure ARHGAP23 GAP activity - PMID:27226243, PMID:28114311 and PMID:32203420 - are absent from it, and none of the three names the gene in its title. This is recorded because passing trust gates certifies that the citations returned are real, not that the decisive ones were found.

Reciprocal arginine-finger test: ARHGAP1's structurally resolved finger (R282, PDB 1TX4) maps onto ARHGAP23 R942, which is ARHGAP23's own annotated Site, and ARHGAP23 R942 maps back onto ARHGAP1 R282. The same holds against ARHGAP21 R1184.

Calibration on controls run through the identical code path: of two proteins known to be GAP-dead, the residue test calls one retained. Retention is therefore not evidence of activity, while loss would have been evidence against it.

At the 25 GAP:GTPase interface positions computed from 1TX4, ARHGAP23 carries the same residue as the experimentally active paralog ARHGAP21 at 22 of them.

Percent interface conservation does not discriminate active from dead in this control set - the GAP-dead ARHGAP11B scores above the experimentally active ARHGAP21 - so it is reported as context and not used as an argument.

The published ARHGAP23-R986K "arginine finger mutant" does not target the annotated arginine finger, and the position it does target does not contact the transition state.