EIF2B3 encodes the gamma subunit of cytosolic eIF2B. Gamma associates with epsilon and helps assemble the exchange-factor complex that regenerates eIF2–GTP for translation initiation. Human tagged-subunit assembly, patient-cell exchange assays and reconstituted human complexes support this contribution. They do not establish that isolated human gamma is an autonomous guanine-nucleotide exchange enzyme. The integrated core therefore records contribution to GEF activity in the eIF2B complex, with cytoplasmic translation initiation and cytosol as its process and functional location. PMID:11323413(https://doi.org/10.1074/jbc.M011788200), PMID:15054402(https://doi.org/10.1038/sj.ejhg.5201189), PMID:15060152(https://doi.org/10.1128/MCB.24.8.3295-3306.2004), PMID:29599213(https://doi.org/10.1126/science.aaq0939).
All 34 imported source contexts are retained: 22 ACCEPT, four MODIFY, four KEEP_AS_NON_CORE and four UNDECIDED. There are no new annotation rows or quotations. Broad translation/initiation assertions are refined to cytoplasmic translation initiation. The gamma–epsilon association is refined to translation initiation factor binding using the human assembly experiments. The campaign rule retains supported generic binding when no justified specific refinement exists; here a supported refinement is available.
The two UniProt IDA GEF assertions associated with PMID:25858979 and PMID:27023709 retain their literal imported enables relation and ACCEPT decisions. Their experiments support the assembled factor's activity. Structured, probable WRONG_QUALIFIER queries ask the source resource to check whether contributes_to is intended; they do not edit the source relation or claim that gamma lacks a functional contribution. Gamma–epsilon assembly evidence is assessed directly, rather than inferred from the roles of other eIF2B subunits.
The human HCV-IRES study supports EIF2B gamma-dependent translation, but its separate EIF2 gamma and ribozyme-cleavage experiments are not evidence of RNA binding by EIF2B gamma. Its Discussion cites an earlier positive RNA-binding observation under the designation eIF-2/2B. The authenticated earlier abstract reports rabbit-reticulocyte competition and binding capacity; the complete preparation, directly contacting factor and gamma contribution could not be resolved. The contributes_to translation-factor/RNA-binding assertion remains UNDECIDED because that complex-level claim remains uncertain, not because it requires isolated gamma binding. PMID:10900014(https://pmc.ncbi.nlm.nih.gov/articles/PMC26988/), PMID:1647211(https://doi.org/10.1016/0167-4781(91)90011-a).
The other unresolved assertions concern the precise heat-response source, gamma-specific oligodendrocyte-development evidence and gamma participation in T-cell-receptor signaling. Positive clinical and whole-factor observations remain recorded. A foregrounded EIF2B5 case or beta/epsilon mechanism is not by itself proof of misattribution. Rat donor records support contextual heat, glucose and peptide-hormone responses; these do not establish direct human gamma sensing. The family-level nervous-system-development inference is retained as non-core.
The current family tree and alignment place the exact 452-residue human sequence below the imported ancestral node. Four literal positive ancestral records support the initiation, GEF and complex annotations. The complete saved ancestral path was checked for loss records; no same-path loss was found. A small donor set and target self-evidence are not defects. The four ARBA rules have satisfied target domain/lineage condition sets. These computational sources are assessed alongside the human assembly evidence, rather than treated as stand-alone experiments.
The three official natural products remain distinct: Q9NR50-1 (452 residues), Q9NR50-2 (412 residues) and Q9NR50-3 (401 residues). Official PDB 6K72 entity 3, chains E/F, and EMDB 9841 macro 3 identify the canonical sequence in a human recombinant complex; substrate eIF2 gamma is a separate entity. Neither those deposits nor a cloned length establishes equal function of the two shorter natural products. PDB 6K72, EMDB 9841.
The source guide distinguishes cached text, separately recovered original sections and unresolved access. Cache metadata is preserved unchanged. Reference-level availability flags describe the recorded access assessment and are omitted where an abstract-only cache coexists with separately read original sections. Supporting references are provided without publication excerpts. Provider acquisition failures did not produce a literature report; the assessment is documented here and in the reproducible source evidence.
The heat-response TAS row is retained as non-core in agreement with the independently supported rat-ortholog transfer. The PMID:12499492 citation remains abstract-only: its exact gamma-specific attribution is unresolved, and the corroborating PMID:9341116 response evidence is not presented as a newly recovered experiment in the clinical paper. Both rows retain their original source fields.
The cytoplasm annotation is refined to cytosol using independent human EIF2B3 localization and the target-resolved Reactome complex. The original PMID:11323413 HeLa fractionation itself resolves cytoplasm only. The electronic CNS-development assignment is marked over-annotated: its matched ARBA conditions establish gamma-family identity, while the cited patient-cell experiment establishes a translation-factor defect rather than a developmental step. The positive human leukoencephalopathy evidence remains intact.
The RNA-binding annotation remains UNDECIDED. This is uncertainty about contribution to the molecular function, not a claim that PMID:10900014 is inaccessible. Its human translation perturbation and the earlier positive eIF-2/2B RNA-binding report do not settle gamma's contribution to that binding function. The contributes_to relation does not require an isolated-gamma binding assay. PMID:1647211 is included as a reference-only support so the earlier abstract and its remaining access limits are directly discoverable.
The resulting 34-row tally is 21 ACCEPT, 5 MODIFY, 4 KEEP_AS_NON_CORE, 3 UNDECIDED and 1 MARK_AS_OVER_ANNOTATED. This supersedes the earlier four-UNDECIDED tally; all three natural products and the single exchange-factor core remain unchanged. No quotation is asserted. Optional supporting_text fields remain absent under the standing zero-new-quotation instruction and the repository's explicit allowance for reference-only evidence.
The two UniProt relation queries remain bounded to the experimentally attributed records and their source-specific assay context. They request CHECK rather than claiming a curator error is proved by complex architecture. The PAINT/ARBA rows preserve their separate propagation provenance; this follow-up does not manufacture additional source-defect submissions merely to make the number of queries symmetrical.
The RNA-binding claim remains UNDECIDED after checking the live GO hierarchy. QuickGO's history records deletion of the GO:0003743 → GO:0008135 is_a relation on 2025-02-12. The current ancestor sets instead place both terms under GO:0180051. Consequently, accepting contribution to initiation-factor activity does not entail contribution to the separate RNA-binding function. This correction concerns the review's ontology premise; the source term and its contributes_to qualifier remain intact, as do the positive observations and the unresolved earlier preparation details.
The exact official ancestor responses and dated relation history are bundled in ontology-review.json, with source URLs and hashes. The 34 actions remain 21 ACCEPT, 5 MODIFY, 4 KEEP_AS_NON_CORE, 3 UNDECIDED and 1 MARK_AS_OVER_ANNOTATED. No publication quotation or new annotation is added.
The source-defect report uses the separately reviewed reference-only renderer correction from PR #5285. Entries without a quotation display their evidence reference as text; no empty quotation or manufactured excerpt is needed.