The primary-source investigation is in TRAF2-deep-research-manual.md. Falcon and Perplexity-lite failed; no provider-generated research is claimed. Receptor/cIAP adaptor function is well grounded. Intrinsic E3 and autoubiquitination are UNDECIDED because primary negative E2-interface evidence conflicts with positive S1P-dependent reconstitution. Complex membership and ubiquitination-process participation are not rejected because of that catalytic uncertainty.
All 708 GOA rows have decisions. Generic interaction rows preserve their individual PMID and partner identifiers, with UNDECIDED where bait/prey evidence was not traced. The full texts of PMID:26458771 and PMID:40097387 explicitly assay TRAF2 despite emphasizing another TRAF in their abstracts; they were read before judging NF-kappaB/TLR and pyroptosis annotations. TNF cytokine binding is distinguished from receptor-complex recovery using the verified GO:0043120 definition. Supporting excerpts in YAML are copied from cached publications.
Remaining follow-up includes the many untraced interaction-map records; the specific TNF ligand-binding annotation; uncommon locations and process contexts; and reconciliation of lipid-cofactor-dependent versus cIAP-mediated ubiquitin transfer. COMPLETE denotes completion of this review pass, not elimination of explicit biological uncertainty.
Scope: full re-review of all 708 GOA rows for Q12933, the description and the core functions. No ProtNLM2 file or prediction-evidence folder was opened or used.
Identity. Q12933 is the reviewed, canonical human TRAF2 product (501 aa, sequence version as in TRAF2-uniprot.txt; RING 34-73, TRAF-type zinc fingers 124-180 and 177-233, coiled coil 299-348, MATH 351-496). It is not a fragment, isoform or alternative gene model. Isoforms 2-4 exist (VSP_007401, VSP_039687, VSP_039688), but no row here is isoform-specific.
Deep research. just deep-research-falcon human TRAF2 --fallback perplexity-lite was retried on 2026-10-20. Falcon failed with "Failed to authenticate"; perplexity-lite returned HTTP 401 (quota exceeded). No provider file was produced. The existing TRAF2-deep-research-manual.md and the cached primary literature were used instead. One extra paper was cached and read: PMID:26701909 (Etemadi et al. 2015 eLife; full text): keratinocyte Traf2 deletion, but not Sphk1 deficiency, disrupts TNF NF-kappaB/MAPK signaling.
Evolutionary grounding. PANTHER PTHR10131:SF21 (TRAF2 orthologs). The PAINT nodes behind the IBAs are: PTN008301857 (signaling adaptor activity; deep TRAF node), PTN000805720 (TNFR binding and TNF signaling; TRAF1/TRAF2 clade), PTN002562154 (Euteleostomi TRAF2 node: ubiquitin-protein transferase activity, positive regulation of canonical NF-kappaB, IRE1-TRAF2-ASK1 complex) and PTN008938480 (cytoplasm; whole family). The mouse Traf2 (P39429) ISS/IEA projections are within the same subfamily, so I accepted the transfer of adaptor-level functions. The family review interpro/panther/PTHR10131/PTHR10131-review.yaml places Q12933 in SF21 and assesses only GO:0035591. I found no disagreement about subfamily placement.
Main changes vs the previous pass.
- Generic protein binding (513 rows): the policy is now applied. 72 rows were changed to MODIFY with a partner-class term: TNFRSF partners to GO:0005164; BIRC2/BIRC3 to GO:0031625; ASK1/NIK/MEKK1 to GO:0031435; IRE1, RIPK1, PKR, RIPK2, TAOK3 and TNIK to GO:0019901; TRADD to GO:0035591. The other 441 rows were changed to REMOVE as uninformative. The REMOVE does not say the interactions are false. The two KEEP_AS_NON_CORE rows (GSTP1, ASK1), which broke the policy, were fixed.
- E3 activity (GO:0004842 and GO:0061630; was UNDECIDED, now KEEP_AS_NON_CORE). For: cIAP-free recombinant TRAF2 has S1P-dependent in vitro E3 activity (PMID:20577214), and PAINT places GO:0004842 at PTN002562154. Against: the RING cannot engage Ubc13-type E2s (PMID:19810754); the TRAF2 RING is dispensable for MAVS signaling, where its E3 activity is redundant with LUBAC (PMID:23951545); and Sphk1 is dispensable in vivo (PMID:26701909). Taken together, the activity is real but not the core of TRAF2 function. Autoubiquitination and sphingolipid binding are also non-core.
- GO:0043120 TNF binding (IPI, partner TRADD) is now REMOVE. TRAF2 is intracellular and cannot bind the cytokine. The same paper's receptor-complex finding is captured by GO:0002947.
- Nucleoplasm (Reactome USP48 reaction) is now MARK_AS_OVER_ANNOTATED. The compartment comes from USP48, and UniProt curates TRAF2 as cytoplasmic.
- Rows whose terms are true but general were changed to MODIFY with the specific term the gene already carries: cytoplasm to cytosol and cytoplasmic side of PM; plasma membrane to cytoplasmic side of PM; signal transduction to TNF-mediated signaling; protein-macromolecule adaptor to signaling adaptor; protein ubiquitination to K63-linked; MAPK cascade to JNK cascade; TNFRSF binding to TNFR binding; signaling receptor complex and protein-containing complex to TNFRSF complex; metal ion to zinc ion; programmed cell death to regulation of apoptotic process; and regulation of production of molecular mediator of immune response to positive regulation of T cell cytokine production.
- Six rows remain UNDECIDED because the evidence cannot be traced: the TLR pathway row (the TIFAB paper uses TRAF2 only as a comparator), cell cortex, vesicle membrane, glial apoptosis, the DNA-damage NAS and the TWEAK IMP.
- The description and core functions were rewritten. Core functions are: signaling adaptor activity in TNFRSF signaling; TNF receptor binding; and cIAP (ubiquitin ligase) binding in K63 ubiquitination and non-canonical NF-kappaB. Suggested questions and experiments were added on whether the RING is needed in vivo.
Relative to review next. Mouse Traf2 (UniProtKB:P39429, MGI:101835) is the ortholog in PTHR10131:SF21 and an experimental donor at PTN002562154, PTN000805720 and PTN008301857. It has extensive knockout data. There is no genes/mouse/Traf2 review, so it needs one.
Modules. TRAF2 appears in modules/tnf_signaling.yaml and modules/nfkb_canonical_signaling.yaml.