ADGRA2

Summary: Well-founded family-level location call, with ADGRA2's own annotation among the seeds.

Reason: PANTHER:PTN001738137 reaches ADGRA1, ADGRA2 and ADGRA3 in human (QuickGO withFrom search, fully paginated, 348 of 348 annotations over 174 entities) and gives them plasma membrane and cell surface receptor signaling pathway - appropriately generic for a node that also spans invertebrate LRR-7TM proteins. The WITH/FROM includes UniProtKB:Q96PE1 itself, a self-referential IBA recording a PAINT curator's judgement that the location is core, which is valid rather than circular.

Summary: Correct but deliberately generic; notable because PAINT chose this parent over GO:0007186 for the whole ADGRA family.

Reason: This is the most informative thing in ADGRA2's IBA record. PANTHER:PTN001738137 covers ADGRA1, ADGRA2 and ADGRA3 in human and 174 entities overall, including invertebrate LRR-7TM proteins, so a generic cell-surface-receptor pathway term is the right LCA for that clade. What matters is what PAINT declined: no ADGRA paralog receives any molecular-function IBA, and none receives GO:0007186 G protein-coupled receptor signaling pathway. PAINT curators looking at this family withheld the GPCR-specific pathway and chose its parent, while InterPro2GO and GDB/TAS assert the GPCR pathway anyway - the two judgements sit side by side in the same GOA record and PAINT's is the better founded. Kept as non-core because the specific pathway ADGRA2 acts in is canonical Wnt signalling.

Summary: Core: ADGRA2 is a component of the Wnt signalosome.

Reason: From PTN002914520, with ADGRA2's own annotation among the WITH/FROM seeds - a valid self-referential IBA recording that a PAINT curator judged the complex membership core. Directly supported in human by the IDA row from PMID:30026314.

Summary: Core: positive regulation of canonical Wnt signalling, placed at the node whose human reach is exactly ADGRA2.

Reason: From PTN002914520 with mouse Adgra2 as a donor, which holds this exact term by IDA (PMID:28803732). Agrees with the human IDA-supported canonical Wnt row and with the Wnt signalosome membership. The propagation lands at the same specificity as its donor, so no downward MODIFY is warranted - the ACRV1 'propagation above its donor' check was run and is negative here.

Summary: True but very broad; ADGRA2's contribution to CNS development is specifically vascular.

Reason: From PTN002914520, whose human reach is exactly ADGRA2, with mouse Adgra2 carrying its own IMP (PMID:21071672). The knockout phenotype is a failure of vessel invasion into neuroepithelium and of cortical expansion, i.e. a vascular defect with a neurodevelopmental consequence. Kept as non-core because the specific processes - sprouting angiogenesis and regulation of blood-brain barrier establishment - are separately annotated and carry the actual biology.

Summary: Sprouting angiogenesis is ADGRA2's best-established organismal role and the IBA sits at the node whose human reach is exactly ADGRA2.

Reason: PANTHER:PTN002914520 was queried directly (QuickGO withFrom search, fully paginated, 136 of 136 annotations over 29 entities): its human reach is ADGRA2 and nothing else, and the four terms it gives are sprouting angiogenesis, CNS development, positive regulation of canonical Wnt signalling and Wnt signalosome - i.e. exactly ADGRA2's characterised biology, correctly placed at the ortholog node rather than at the ADGRA family node. Both non-self donors are true adgra2 orthologs and mouse Adgra2 carries its own IMP for this term. The affinage deep-research record is cited here only as a phenotype summary, not for any mechanistic claim: every element of it - embryonic lethality, forebrain and spinal cord angiogenesis defects, failure of vessel invasion into the neuroepithelium, loss of barrier properties, impaired cortical expansion, and the endothelial-specific knockout phenocopying the global one - was checked against the cached full text of PMID:21421844 before being used.

Summary: Generic transmembrane-receptor term derived from the family-2 GPCR signature; defensible but not the characterised activity.

Reason: IPR017981 is a 'GPCR, family 2-like' signature and ADGRA2 genuinely has that seven-transmembrane architecture, so unlike its child GO:0004930 this term makes no claim about G-protein coupling and is not refuted. It is not accepted either, for a reason worth stating explicitly: GO:0004888 requires the protein to combine with a signal and transmit it across the membrane, and ADGRA2 does not bind Wnt7 - RECK does. Whether ADGRA2 satisfies the definition therefore turns on whether the RECK-bound Wnt7 complex counts as 'the signal' it combines with. That is a genuine ambiguity rather than a settled yes or no, and it is what KEEP_AS_NON_CORE is for. The informative molecular function, and the one that needs no such reading, is the adaptor activity proposed as a NEW row (GO:0030159).

Summary: G protein-coupled receptor activity is asserted from structural signatures alone; ADGRA2 has no known ligand and no measured G-protein coupling.

Reason: GO:0004930 requires that the protein combine with an extracellular signal and promote GDP/GTP exchange on a heterotrimeric G-protein alpha subunit. None of the three requirements has been met for ADGRA2. It is an orphan: IUPHAR/GtoPdb target 198 returns no natural ligand and an empty interactions list, and records no transducer. No coupling assay has been reported for ADGRA2 itself - and this now rests on a published statement rather than on the absence of search hits: the 2025 study that did measure coupling in this clade says so directly, 'Whereas ADGRA2 has never been shown to couple to heterotrimeric G proteins, it has been extensively studied for its role in the assembly of a complex signalosome' (PMID:40127866). 'GPR124 Stachel' and 'GPR124 tethered agonist' return zero PubMed records, an ADGRA2-specific negative. The single reported heterotrimeric-G-protein contact runs the opposite way to the definition: Gbeta-gamma binds the ADGRA2 C-tail and promotes assembly of a GPR124-Elmo complex. UniProt's own FUNCTION block states that the characterised ADGRA2 tethering activity does not rely on its GPCR structure. PAINT agrees: across all three human ADGRA paralogs there is no molecular-function IBA at all. Marked over-annotated rather than removed, and the sibling paralogs are what make that the right call rather than a hedge. This clade demonstrably CAN couple: ADGRA3 shows low-level activation of Gi and Gs, more from its C-terminal fragment, abrogated when the first three residues of the stachel tethered agonist are removed (PMID:40127866), and ADGRA1 activates several G proteins, notably G-alpha-13, on the full TRUPATH BRET2 panel (PMID:41961591). So a fold-based inference of coupling is not absurd for an ADGRA protein - it is simply unverified for this member, whose 7TM plus GAIN-B architecture is genuine and has never been assayed and failed. An unmeasured claim, not a refuted one. Naming the responsible signatures makes the defect specific rather than generic, and the authoritative interpro2go mapping shows InterPro curators already drew the distinction being argued for here. IPR017981 'GPCR, family 2-like, 7TM' is the entry that covers ADGRA2's transmembrane bundle, and InterPro describes it as representing the transmembrane domain of family 2 GPCRs AND Frizzled proteins - Frizzleds being the textbook 7TM family whose principal signalling is not through heterotrimeric G proteins. Accordingly interpro2go maps IPR017981 only to the generic GO:0004888 and GO:0007166, never to GO:0004930 or GO:0007186, and that is exactly which terms ADGRA2 receives from it. The GPCR-activity claim comes instead from four secretin-receptor-specific entries: IPR000832 and IPR017983 (secretin-like family 2 and its conserved site) and IPR001879 with its superfamily IPR036445, which are not the GPS or GAIN entries but the family-2 extracellular HORMONE RECEPTOR domain - InterPro calls it 'the major ligand recognition domain' and names calcitonin, corticotropin-releasing-factor 1, diuretic hormone, GLP-1 and parathyroid hormone receptors as its exemplars. So a peptide-hormone ligand-recognition signature, matched on ADGRA2's GAIN/hormone-receptor module, is being converted into 'activates a G protein' for a receptor that has no ligand at all. That is a sharper objection than a generic fold-to-activity complaint, and it is testable: the one signature InterPro2GO already restricts is the one whose membership includes non-coupling 7TMs.

Summary: Combined automatic pipeline call agreeing with the experimental rows.

Reason: Derived from ARBA plus the mouse ortholog plus the UniProt SubCell keyword SL-0039; all three agree with the direct human evidence.

Summary: Generic pathway term from the family-2 GPCR signature; true but uninformative.

Reason: IPR017981 supports membership of a cell-surface receptor family and this term makes no G-protein claim, so it survives where the GO:0007186 row from the same pipeline does not.

Summary: GPCR signalling pathway asserted from secretin-like 7TM signatures; no ligand, no measured G-protein coupling, and PAINT withheld this term from the whole ADGRA family.

Reason: GO:0007186 requires a ligand-bound receptor to promote GDP/GTP exchange on an associated G-alpha subunit and the activated G-alpha to dissociate from G-beta-gamma. GO:0004930 requires that the protein combine with an extracellular signal and promote GDP/GTP exchange on a heterotrimeric G-protein alpha subunit. None of the three requirements has been met for ADGRA2. It is an orphan: IUPHAR/GtoPdb target 198 returns no natural ligand and an empty interactions list, and records no transducer. No coupling assay has been reported for ADGRA2 itself - and this now rests on a published statement rather than on the absence of search hits: the 2025 study that did measure coupling in this clade says so directly, 'Whereas ADGRA2 has never been shown to couple to heterotrimeric G proteins, it has been extensively studied for its role in the assembly of a complex signalosome' (PMID:40127866). 'GPR124 Stachel' and 'GPR124 tethered agonist' return zero PubMed records, an ADGRA2-specific negative. The single reported heterotrimeric-G-protein contact runs the opposite way to the definition: Gbeta-gamma binds the ADGRA2 C-tail and promotes assembly of a GPR124-Elmo complex. UniProt's own FUNCTION block states that the characterised ADGRA2 tethering activity does not rely on its GPCR structure. PAINT agrees: across all three human ADGRA paralogs there is no molecular-function IBA at all. PAINT reached the same conclusion independently: the ADGRA family node PTN001738137 gives ADGRA1, ADGRA2 and ADGRA3 the parent term GO:0007166 and never GO:0007186. Marked over-annotated rather than removed, and the sibling paralogs are what make that the right call rather than a hedge. This clade demonstrably CAN couple: ADGRA3 shows low-level activation of Gi and Gs, more from its C-terminal fragment, abrogated when the first three residues of the stachel tethered agonist are removed (PMID:40127866), and ADGRA1 activates several G proteins, notably G-alpha-13, on the full TRUPATH BRET2 panel (PMID:41961591). So a fold-based inference of coupling is not absurd for an ADGRA protein - it is simply unverified for this member, whose 7TM plus GAIN-B architecture is genuine and has never been assayed and failed. An unmeasured claim, not a refuted one. Naming the responsible signatures makes the defect specific rather than generic, and the authoritative interpro2go mapping shows InterPro curators already drew the distinction being argued for here. IPR017981 'GPCR, family 2-like, 7TM' is the entry that covers ADGRA2's transmembrane bundle, and InterPro describes it as representing the transmembrane domain of family 2 GPCRs AND Frizzled proteins - Frizzleds being the textbook 7TM family whose principal signalling is not through heterotrimeric G proteins. Accordingly interpro2go maps IPR017981 only to the generic GO:0004888 and GO:0007166, never to GO:0004930 or GO:0007186, and that is exactly which terms ADGRA2 receives from it. The GPCR-activity claim comes instead from four secretin-receptor-specific entries: IPR000832 and IPR017983 (secretin-like family 2 and its conserved site) and IPR001879 with its superfamily IPR036445, which are not the GPS or GAIN entries but the family-2 extracellular HORMONE RECEPTOR domain - InterPro calls it 'the major ligand recognition domain' and names calcitonin, corticotropin-releasing-factor 1, diuretic hormone, GLP-1 and parathyroid hormone receptors as its exemplars. So a peptide-hormone ligand-recognition signature, matched on ADGRA2's GAIN/hormone-receptor module, is being converted into 'activates a G protein' for a receptor that has no ligand at all. That is a sharper objection than a generic fold-to-activity complaint, and it is testable: the one signature InterPro2GO already restricts is the one whose membership includes non-coupling 7TMs.

Summary: Redundant parent of plasma membrane, which ADGRA2 already holds by direct human evidence.

Reason: Not wrong - ADGRA2 is a multi-pass membrane protein - but GO:0005886 is supported here by three EXP rows, an IDA and an IBA, so the generic parent adds nothing.

Summary: Automatic transfer of UniProt's subcellular-location line, which itself carries experimental support.

Reason: The SubCell term SL-0286 maps from UniProt's 'Cell projection, filopodium' assignment, which is tagged ECO:0000269|PubMed:21421844 - so this IEA faithfully reflects an experimentally-supported curator statement rather than a prediction. Independently corroborated by PMID:28600358, which reports enrichment of GPR124 at actin-rich protrusions containing the filopodial marker VASP. Non-core: a sub-localisation of the plasma-membrane pool, not the gene's function.

Summary: ARBA machine-learning call that happens to agree with the direct human evidence.

Reason: Redundant with the IDA row from PMID:30026314 but not wrong.

Summary: ARBA call agreeing with the human IDA.

Reason: Redundant with the IDA row from PMID:30026314.

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open. Partner-specific note: DLG1 (SAP97), 3 PDZ domains. The one partner with dedicated non-screen evidence: PMID:15021905 mapped the interaction to the hDlg PDZ domains and the TEM5 C-terminal PDZ-binding motif, and UniProt records it in SUBUNIT.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open. Partner-specific note: DLG1 splice isoform 2 (926 aa vs 904 aa canonical) - a genuine UniProt isoform, not a partial clone.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open. Partner-specific note: DLG1 (SAP97), 3 PDZ domains. The one partner with dedicated non-screen evidence: PMID:15021905 mapped the interaction to the hDlg PDZ domains and the TEM5 C-terminal PDZ-binding motif, and UniProt records it in SUBUNIT.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open. Partner-specific note: DLG1 (SAP97), 3 PDZ domains. The one partner with dedicated non-screen evidence: PMID:15021905 mapped the interaction to the hDlg PDZ domains and the TEM5 C-terminal PDZ-binding motif, and UniProt records it in SUBUNIT.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Bare protein binding replaced by the specific measured activity: this row is one PDZ domain scored against ADGRA2's C-terminal PDZ-binding motif.

Reason: Bare protein binding is uninformative, and here it can be replaced by a specific, well-measured activity. All 21 GO:0005515 rows on ADGRA2 come from assays of one element - the C-terminal class-I PDZ-binding motif ...LWKSETTV - against PDZ domains. Resolving every WITH/FROM accession from the GOA TSV gives 19 distinct partners, 19/19 reviewed Swiss-Prot at canonical length (no TrEMBL or ORFeome substitutions), and 18/19 carrying at least one annotated PDZ domain; the exception is Q12959-2, a genuine DLG1 splice isoform whose features are not returned per-isoform. GO:0030165 PDZ domain binding captures exactly what was measured and is honest about a motif whose binding is intrinsically promiscuous, whereas naming 19 individual partners as biological interactors would over-read a domain-affinity profile. Not promoted to a core function: the biochemistry is solid but the mammalian physiology is not. PMID:35649360 reports that mammalian GPR124 engages Frizzled through an ICD-independent modality, so the in vivo role of the conserved human ETTV motif and its PDZ partners is open.

Proposed replacements: PDZ domain binding

Summary: Ensembl Compara transfer from mouse Adgra2, which holds this term by its own IMP.

Reason: Donor UniProtKB:Q91ZV8 is the true 1:1 mouse ortholog and carries GO:0002040 by IMP (PMID:23918385, PMID:21071672), so the transfer is not circular and not a family-level guess.

Summary: Compara transfer of the same broad developmental term from mouse Adgra2.

Reason: Donor holds GO:0007417 by IMP (PMID:21071672); the term is correct but too broad to be core, for the same reason as the IBA row.

Summary: Compara transfer from mouse Adgra2, which holds the term by IMP.

Reason: Donor UniProtKB:Q91ZV8 carries GO:0043542 by IMP (PMID:21071672). Endothelial migration is a genuine downstream consequence of ADGRA2 function but is not its molecular role, so non-core. Note the donor also carries GO:0010595 positive regulation of endothelial cell migration by IDA, which was not transferred; that term is proposed as a NEW row below.

Summary: Compara transfer from mouse Adgra2, which holds the term by IMP.

Reason: Correct but largely redundant with GO:0002040 sprouting angiogenesis, which is the specific process the knockout phenotype identifies.

Summary: Compara transfer from mouse Adgra2 (IMP PMID:21071672); a peripheral description of the endothelial guidance defect.

Reason: The underlying mouse phenotype is directional endothelial migration into the neuroepithelium. 'Regulation of chemotaxis' is a defensible but oblique way to describe it, and no chemoattractant for ADGRA2 has been identified - the receptor is an orphan. Kept rather than removed because the donor annotation is a curator's experimental call, but it should not be read as evidence that ADGRA2 responds to a chemotactic ligand.

Summary: Core organismal role: ADGRA2 is required for blood-brain barrier establishment.

Reason: Donor mouse Adgra2 holds this term by two independent IMPs (PMID:21421844, PMID:28288111). Function and phenotype are kept separate here: the knockout establishes the biological process, while the molecular activity underlying it is the Wnt7 co-activation annotated separately.

Summary: Compara transfer of the same term from mouse Adgra2 (IDA PMID:28803732).

Reason: Redundant with the IBA row and with the human IDA-supported canonical Wnt annotation.

Summary: HPA immunofluorescence places ADGRA2 at the plasma membrane, agreeing with three independent experimental papers.

Reason: Concordant with the EXP rows from PMID:16982628, PMID:21421844 and PMID:22013897 and with UniProt's Cell membrane assignment.

Summary: Cell-surface expression of TEM5/ADGRA2 on endothelial cells shown directly.

Reason: Human endothelial cells; UniProt records the same paper as experimental support for the Cell membrane location (ECO:0000269|PubMed:16982628).

Summary: Plasma-membrane localisation supported by the GPR124 knockout and localisation study.

Reason: UniProt cites this paper with ECO:0000269 for both Cell membrane and Cell projection/filopodium, and adds that the receptor is enriched at lateral cell borders and at sites of cell-ECM contact. The cached full text is partial and does not contain the localisation figure, so the curator's reading is deferred to rather than second-guessed.

Summary: Cell-surface ADGRA2 is the substrate for thrombin-induced ectodomain shedding, which requires cell-surface PDI.

Reason: Shedding of the N60 fragment from intact cells, and its dependence on cell-surface protein disulfide-isomerase, is only interpretable if the receptor is at the plasma membrane.

Summary: Core: ADGRA2 is a Wnt7-specific co-activator of canonical Wnt/beta-catenin signalling.

Reason: ADGRA2 is a Wnt7-specific co-activator: RECK captures WNT7A/WNT7B and ADGRA2 makes the RECK-bound ligand available to Frizzled by nucleating a higher-order RECK-ADGRA2-FZD-LRP5/6 signalosome. This is the gene's defining molecular role and is independently supported by PMID:28289266, which places ADGRA2 with RECK as the WNT7B co-receptors required for synergistic beta-catenin signalling alongside FZD5, FZD8 and LRP6.

Summary: Sequence-similarity transfer from mouse Adgra2, which holds the term by EXP and IDA.

Reason: Donor UniProtKB:Q91ZV8 carries GO:0005886 by EXP (PMID:25558062) and IDA (PMID:28803732); the human protein also has its own direct evidence, so the row is redundant rather than load-bearing.

Summary: Curator sequence-similarity transfer of the BBB role from mouse Adgra2.

Reason: Same donor with two mouse IMPs; WITH/FROM correctly names only the sequence-similar entity.

Summary: Core: direct demonstration that ADGRA2 is recruited into Wnt/Frizzled/LRP5/6 signalosomes.

Reason: ADGRA2 is a Wnt7-specific co-activator: RECK captures WNT7A/WNT7B and ADGRA2 makes the RECK-bound ligand available to Frizzled by nucleating a higher-order RECK-ADGRA2-FZD-LRP5/6 signalosome. The complex is the physical basis of the co-activator function, and GO:1990909's own definition (a multiprotein complex containing membrane-localized Wnt receptors and cytosolic complexes, with at least a Wnt protein, LRP5/6, a Frizzled, Axin and Dishevelled) matches the assembly described.

Summary: Curator sequence-similarity transfer of the broad developmental term from mouse Adgra2.

Reason: Same donor and same mouse IMP; correct but not core.

Summary: Curator transfer of the same term from the same mouse donor.

Reason: Same donor and evidence as the Compara row.

Summary: Curator transfer of the same term from the same mouse donor.

Reason: Same donor and evidence as the Compara row; same caveat about the absence of any identified chemoattractant.

Summary: Curator sequence-similarity transfer from mouse Adgra2; donor holds the term experimentally.

Reason: Same donor and same underlying mouse IMP evidence as the Compara row; WITH/FROM correctly lists only the sequence-similar entity, as ISS requires.

Summary: Curator transfer of the general angiogenesis-regulation term from mouse Adgra2.

Reason: Same donor and mouse IMP; redundant with the specific sprouting angiogenesis annotation.

Summary: TAS to a repertoire-cataloguing paper that cannot supply a gene-specific author statement about ADGRA2's activity; the same reference gives 27 human entities the same generic triple, two of them pseudogenes.

Reason: GO:0004930 requires that the protein combine with an extracellular signal and promote GDP/GTP exchange on a heterotrimeric G-protein alpha subunit. None of the three requirements has been met for ADGRA2. It is an orphan: IUPHAR/GtoPdb target 198 returns no natural ligand and an empty interactions list, and records no transducer. No coupling assay has been reported for ADGRA2 itself - and this now rests on a published statement rather than on the absence of search hits: the 2025 study that did measure coupling in this clade says so directly, 'Whereas ADGRA2 has never been shown to couple to heterotrimeric G proteins, it has been extensively studied for its role in the assembly of a complex signalosome' (PMID:40127866). 'GPR124 Stachel' and 'GPR124 tethered agonist' return zero PubMed records, an ADGRA2-specific negative. The single reported heterotrimeric-G-protein contact runs the opposite way to the definition: Gbeta-gamma binds the ADGRA2 C-tail and promotes assembly of a GPR124-Elmo complex. UniProt's own FUNCTION block states that the characterised ADGRA2 tethering activity does not rely on its GPCR structure. PAINT agrees: across all three human ADGRA paralogs there is no molecular-function IBA at all. The cited reference is a genome-mining and phylogenetics survey. Its abstract describes genome-database searching, phylogenetic analysis and EST expression charting and reports no functional assay on any receptor; the paper is abstract-only in the cache, so nothing here rests on a claim about its full text. Querying QuickGO by reference rather than by gene (?reference=PMID:15203201, fully paginated, 78 of 78 rows read) returns 78 annotations over 27 distinct human entities, 100% TAS and 100% assignedBy GDB, carrying one identical generic triple: GO:0016020 to 27 entities, GO:0007186 to 26 and GO:0004930 to 25. The two entities that miss GO:0004930 are ADGRG3 and ADGRV1, both protein-coding, so the pseudogene recipients ADGRE4P and ADGRF2P are confirmed to be among the 25 that DO receive the molecular function - checked by name rather than inferred from the 25-of-27 count, which on its own is equally consistent with the pseudogenes being the two that missed it. One reference giving 27 entities the same triple with identical evidence is a bulk classification import, not 27 traceable author statements. The whole argument is a statement about the DISTRIBUTION of annotations and needs no access to the full text: a paper cannot make 25 gene-specific author statements about G-protein coupling, two of them about pseudogenes, as a by-product of cataloguing a repertoire. Reproducible via ADGRA2-bioinformatics/projection_test.py -> projection_test.json. Action calibrated deliberately against the InterPro2GO rows carrying the same term, because the two routes fail differently. The InterPro rows rest on a signature that genuinely matches ADGRA2, so the inference has a real if over-extended basis and is marked over-annotated. This row rests on an evidence code that asserts a traceable author statement about ADGRA2, and the reference cannot supply one. The distribution analysis above is what establishes that, and it needs no access to the full text. That is a defect in the annotation regardless of whether the underlying claim might one day be demonstrated, which is why REMOVE here is not in tension with the finding that the claim itself is unmeasured rather than refuted: the InterPro rows remain, so removing this one does not erase the claim from GOA. Removing two rows on this gene is also complementary to, not a substitute for, the suggested_questions item asking GO whether the whole 78-annotation GDB block across 27 entities should be retired - that is a consortium-level decision, this is the gene-level one a gene review exists to make. For symmetry, note what is NOT removed: the GO:0016020 TAS row from this same GDB block carries the identical evidentiary defect and would be a legitimate REMOVE on the same reasoning. It is kept because removing it would achieve nothing - membrane localisation is directly demonstrated for ADGRA2 elsewhere in this very GOA record, by three EXP rows, an IDA and an IBA to the child term GO:0005886 - so that row is redundant rather than misleading. Its shared defect is recorded in its own review and in the block-level suggested_questions item rather than acted on by deleting a redundantly-supported true statement. Cross-review consistency, stated explicitly because the sibling review reaches the opposite action on the identical row. ADGRA3 (merged) ACCEPTs both PMID:15203201 TAS rows - 'right conclusion, weak provenance - kept because the conclusion has since been earned' - and recommends GOA re-evidence them against PMID:40127866. That is not an inconsistency with this REMOVE; both follow one rule: a defective TAS row is kept where the claim it makes is independently established for that gene, and removed where nothing else on the gene carries it. For ADGRA3 the claim was earned by a direct Gi/Gs coupling measurement. For ADGRA2 it was not, and the same paper says so. The rule is visible inside this review too, which is the check that it is a rule and not a rationalisation: the GO:0016020 TAS row from this very block is kept on exactly ADGRA3's logic, because plasma-membrane localisation IS independently demonstrated for ADGRA2. Flagged here rather than left for someone to notice later, per the AADACL2/3/4 precedent where three reviews gave one row three answers and nobody reconciled them.

Summary: Same GDB bulk block as the GO:0004930 TAS row: one repertoire paper, 27 human entities, one generic triple.

Reason: GO:0007186 is a process term with its own requirement, distinct from the molecular function: a ligand must bind the receptor, the activated receptor must promote GDP/GTP exchange on an associated G-alpha, and the GTP-bound G-alpha must then dissociate from G-beta-gamma to transmit the signal onward. For ADGRA2 the pathway fails at its first step as well as its second - no ligand is known - and no downstream G-protein-dependent readout has been attributed to ADGRA2 in any system. The cited reference is a genome-mining and phylogenetics survey. Its abstract describes genome-database searching, phylogenetic analysis and EST expression charting and reports no functional assay on any receptor; the paper is abstract-only in the cache, so nothing here rests on a claim about its full text. Querying QuickGO by reference rather than by gene (?reference=PMID:15203201, fully paginated, 78 of 78 rows read) returns 78 annotations over 27 distinct human entities, 100% TAS and 100% assignedBy GDB, carrying one identical generic triple: GO:0016020 to 27 entities, GO:0007186 to 26 and GO:0004930 to 25. The two entities that miss GO:0004930 are ADGRG3 and ADGRV1, both protein-coding, so the pseudogene recipients ADGRE4P and ADGRF2P are confirmed to be among the 25 that DO receive the molecular function - checked by name rather than inferred from the 25-of-27 count, which on its own is equally consistent with the pseudogenes being the two that missed it. One reference giving 27 entities the same triple with identical evidence is a bulk classification import, not 27 traceable author statements. The whole argument is a statement about the DISTRIBUTION of annotations and needs no access to the full text: a paper cannot make 25 gene-specific author statements about G-protein coupling, two of them about pseudogenes, as a by-product of cataloguing a repertoire. Reproducible via ADGRA2-bioinformatics/projection_test.py -> projection_test.json. GO:0004930 requires that the protein combine with an extracellular signal and promote GDP/GTP exchange on a heterotrimeric G-protein alpha subunit. None of the three requirements has been met for ADGRA2. It is an orphan: IUPHAR/GtoPdb target 198 returns no natural ligand and an empty interactions list, and records no transducer. No coupling assay has been reported for ADGRA2 itself - and this now rests on a published statement rather than on the absence of search hits: the 2025 study that did measure coupling in this clade says so directly, 'Whereas ADGRA2 has never been shown to couple to heterotrimeric G proteins, it has been extensively studied for its role in the assembly of a complex signalosome' (PMID:40127866). 'GPR124 Stachel' and 'GPR124 tethered agonist' return zero PubMed records, an ADGRA2-specific negative. The single reported heterotrimeric-G-protein contact runs the opposite way to the definition: Gbeta-gamma binds the ADGRA2 C-tail and promotes assembly of a GPR124-Elmo complex. UniProt's own FUNCTION block states that the characterised ADGRA2 tethering activity does not rely on its GPCR structure. PAINT agrees: across all three human ADGRA paralogs there is no molecular-function IBA at all. Action calibrated deliberately against the InterPro2GO rows carrying the same term, because the two routes fail differently. The InterPro rows rest on a signature that genuinely matches ADGRA2, so the inference has a real if over-extended basis and is marked over-annotated. This row rests on an evidence code that asserts a traceable author statement about ADGRA2, and the reference cannot supply one. The distribution analysis above is what establishes that, and it needs no access to the full text. That is a defect in the annotation regardless of whether the underlying claim might one day be demonstrated, which is why REMOVE here is not in tension with the finding that the claim itself is unmeasured rather than refuted: the InterPro rows remain, so removing this one does not erase the claim from GOA. Removing two rows on this gene is also complementary to, not a substitute for, the suggested_questions item asking GO whether the whole 78-annotation GDB block across 27 entities should be retired - that is a consortium-level decision, this is the gene-level one a gene review exists to make. For symmetry, note what is NOT removed: the GO:0016020 TAS row from this same GDB block carries the identical evidentiary defect and would be a legitimate REMOVE on the same reasoning. It is kept because removing it would achieve nothing - membrane localisation is directly demonstrated for ADGRA2 elsewhere in this very GOA record, by three EXP rows, an IDA and an IBA to the child term GO:0005886 - so that row is redundant rather than misleading. Its shared defect is recorded in its own review and in the block-level suggested_questions item rather than acted on by deleting a redundantly-supported true statement. Cross-review consistency, stated explicitly because the sibling review reaches the opposite action on the identical row. ADGRA3 (merged) ACCEPTs both PMID:15203201 TAS rows - 'right conclusion, weak provenance - kept because the conclusion has since been earned' - and recommends GOA re-evidence them against PMID:40127866. That is not an inconsistency with this REMOVE; both follow one rule: a defective TAS row is kept where the claim it makes is independently established for that gene, and removed where nothing else on the gene carries it. For ADGRA3 the claim was earned by a direct Gi/Gs coupling measurement. For ADGRA2 it was not, and the same paper says so. The rule is visible inside this review too, which is the check that it is a rule and not a rationalisation: the GO:0016020 TAS row from this very block is kept on exactly ADGRA3's logic, because plasma-membrane localisation IS independently demonstrated for ADGRA2. Flagged here rather than left for someone to notice later, per the AADACL2/3/4 precedent where three reviews gave one row three answers and nobody reconciled them.

Summary: Same GDB bulk block as the two removed GPCR rows, carrying the identical evidentiary defect; kept only because the term is redundantly and directly supported elsewhere on this gene.

Reason: Leading with the defect rather than with the truth of the claim, because the defect is what this row shares with the two that were removed. It is one of 27 identical GO:0016020 TAS annotations that PMID:15203201 places on 27 adhesion-GPCR entities including two pseudogenes, so the TAS code overstates what the reference supplies here exactly as it does for GO:0004930 and GO:0007186, and on evidentiary grounds alone this row would be a legitimate REMOVE too. It is kept because removing it would achieve nothing: membrane localisation is directly demonstrated for ADGRA2 elsewhere in this same GOA record, by three EXP rows (PMID:16982628, PMID:21421844, PMID:22013897), an IDA from HPA and an IBA, all to the child term GO:0005886. The operative rule is remove where the defective row is carrying the claim, flag where it merely duplicates a directly demonstrated one - and this row only duplicates. Non-core because GO:0005886 already supplies the specific location; the shared provenance defect is escalated in the block-level suggested_questions item rather than acted on by deleting a redundantly-supported true statement.

Summary: ADGRA2 has no informative molecular-function annotation; this is the activity the Wnt7 literature actually establishes.

Reason: ADGRA2's GOA molecular functions are bare protein binding plus two unsupported GPCR terms, so the gene's characterised activity is unrepresented. GO:0030159 is defined as 'the binding activity of a molecule that provides a physical support for the assembly of a multiprotein receptor signaling complex', which is precisely what ADGRA2 does: ADGRA2 is a Wnt7-specific co-activator: RECK captures WNT7A/WNT7B and ADGRA2 makes the RECK-bound ligand available to Frizzled by nucleating a higher-order RECK-ADGRA2-FZD-LRP5/6 signalosome. UniProt states the same, and adds that the activity does not depend on the GPCR structure but on binding RECK extracellularly and Dishevelled intracellularly. The term is already used this way for co-receptors and receptor scaffolds - human CD19, a B-cell co-receptor, holds it by EXP, and CDH5, LAT, IRS1 and MAGI2 hold it too. On species scope: the suggested_questions item about PMID:30026314 concerns WHICH DOMAIN mediates the interaction, not whether ADGRA2 performs the adaptor role, and the two must not be conflated - PMID:35649360 contrasts zebrafish, mouse and human Gpr124 explicitly and concludes the mammalian receptors use an alternative intracellular-domain-independent modality, which relocates the mechanism rather than withdrawing the function. The row is additionally supported by PMID:28289266, an independent group; its abstract reports the effect 'in multiple cell types' without naming them. What cannot be stated from the cached records is the cell type or species used for the signalosome imaging in PMID:30026314, because that paper and PMID:28289266 are both abstract-only here and neither abstract names it; that is recorded as unverified rather than assumed.

Summary: A direct, human, experimentally-demonstrated interaction that UniProt records and GO does not have at all.

Reason: The ADGRA2 ectodomain is shed and, after further proteolytic processing, its cryptic RGD motif (UniProt FT MOTIF 362..364) engages integrin alpha-V/beta-3 directly. UniProt records this in both SUBUNIT ('The cleaved extracellular subunit interacts with the integrin heterodimer ITGAV:ITGB3') and DOMAIN ('The RGD motif is involved in integrin ITGAV:ITGB3 binding'), both tagged ECO:0000269|PubMed:16982628. No GO:0005178 row exists on ADGRA2. The interaction requires prior shedding and processing, which should be recorded in the annotation extension or comment rather than being read as a property of the intact receptor.

Summary: The shed ectodomain binds several glycosaminoglycans; UniProt records it and GO has no corresponding row.

Reason: UniProt's FUNCTION block states that ADGRA2 'Binds to the glycosaminoglycans heparin, heparin sulfate, chondroitin sulfate and dermatan sulfate (PubMed:16982628)'. GO:0005539 glycosaminoglycan binding, rather than the narrower GO:0008201 heparin binding, is the correct term because all four classes were bound - choosing heparin alone would assert a specificity the data do not show. Together with the integrin interaction this is the bridging activity named in the source paper's own title.

Summary: Completeness gap: the mouse donor holds this term by IDA but only its unregulated parent was transferred to human.

Reason: Querying the ISS/Compara donor's own GOA shows mouse Adgra2 (UniProtKB:Q91ZV8) carries GO:0010595 by IDA from PMID:21421844, alongside the GO:0043542 IMP that did transfer. Because the donor holds both, this is a transfer-set completeness gap rather than a propagation that landed above its donor. Proposed as ISS with the sequence-similar entity alone in WITH/FROM, as ISS semantics require. The donor additionally holds GO:0009986 cell surface (IDA), GO:0001525 angiogenesis (IMP) and GO:1900747 negative regulation of VEGF signaling pathway (IMP), none transferred; those are flagged in suggested_questions rather than proposed here, because the human evidence for them is thinner.