The evidence archive preserves the 279 original annotation objects, their forecast/native joins, the three extension atoms, product identities, ancestral assertions and interaction-source locators. Row numbers below are zero-based original raw indices. Positive interaction records are distinct from the decision to retain an informative GFAP molecular function.
The human sequence and intermediate-filament architecture are documented in PMID2740350, whose original scans were unavailable. PMID35511821 directly tests purified recombinant human wild-type GFAP: dialysis, electron microscopy and sedimentation support filament assembly. Human-protein expression in rat astrocytes is a separate cellular experiment. Selected original assembly Methods/Results were read; images and supplements were not reanalyzed.
PMID12058025 was read in selected original Methods, Results and captions. Human alpha and epsilon entered cellular filament networks, allowing heteromeric incorporation. Epsilon-tail binding involved presenilin fragments, not proof of a whole native complex. Engineered GFAP46/deletion reagents are not additional natural products.
The authentic PMID17203480 abstract reports kappa homomer failure and disruption of alpha networks as kappa abundance rises. Pig developmental transcript ratios and proposed motility effects are separate. Full text was unavailable.
The official inventory is alpha P14136-1, 432 residues; epsilon P14136-3, 431 residues; and kappa P14136-2, 438 residues. Display names 2 and 3 therefore do not match accession suffixes. All three direct FASTAs agree with the official sequence edits. Functional splice classes retain these mappings; no product-exclusive GO terms are invented.
PMID6371809 was read in all five original scanned pages. Human glioma immunogold labeling places GFAP and vimentin in the same filaments; specificity controls support the identities. The result does not resolve a particular heterodimer or exclude separate protofibrils. This supplies functional context for the vimentin binding refinements. NEFL and KRT13 screen/PCA results do not independently establish comparable GFAP filament roles.
PMID20797626 places GFAP on the cytosolic lysosomal surface in mouse/rat experiments. LAMP2A stabilization opposes Hsc70-driven disassembly; an EEF1A1-associated pool favors GFAP self-association. Initial complexes form without GFAP. Purified-reagent sheep identity remains unresolved. The PMID20176123 review described this work while under revision.
For PMID26571399, selected original text and all supplementary pages were inspected. The exact GFAP experiment behind the two IDAs remains unresolved. Col6a4 perturbation and glial-marker measurements do not justify declaring misattribution. The native cache flag is misleadingly broad: its text omits Results and Methods. Both annotations remain undecided with all literal context preserved.
Other donor evidence remains bounded: 16988027 supplies rat GFAP–integrin co-immunoprecipitation and mouse corroboration, not purified human contact; 11082283 explicitly reports Nm23 association in rat C6 cells; 21250919 separates GFAP knockdown effects on C6 proliferation from combined injury treatments. The 32057593 rat ethanol endpoint is increased Gfap transcript, not proof that GFAP executes ethanol-response biology.
Positive interactions and independent partner classifications remain preserved. Neither a credible interaction nor a correct partner class alone supplies a useful GFAP molecular function. The follow-up removes 134 generic binding annotations for informativeness; it does not classify them as artifacts. Four vimentin refinements remain because binding converges with independently demonstrated GFAP structural function. Binding terms do not require GFAP to activate its partner or perform its downstream function. Experimental, curated-family and similarity grades remain distinct in the archive; the earlier class-based judgments are historical provenance.
| Source | Original evidence and limit |
|---|---|
| 16189514 | Five exact pairs in the author-maintained H-I-05 release, two Y2H method records each. This later release is distinguished from unrecovered original S2/S3. GFAP ORF2920 mapping is not independent original-clone sequencing. |
| 25416956 | Original S2G contains 17 heterotypic pairs and the self pair. Individual quadruplicate retesting and autoactivation controls support the final screen set. Available canonical-clone sequence does not add a per-pair isoform field. |
| 25910212 | Seven positive wild-type pairs occur at S3A rows1013–1019; none appears in the GFAP GPCA subset. The target CDS exactly translates to 432-residue canonical GFAP. PIAS2 remains a gene-level partner assignment. |
| 26871637 | Both NEFL pairs are positive. NEFL_1 exactly matches I6L9F6; published NEFL_2 sequence establishes P07196 residues1–377 plus one trailing nucleotide. This incomplete table sequence does not prove the physical construct was truncated. Raw PCA scores are retained without an invented threshold. |
| 32296183 | The 132 literal HuRI target records retain assay, ORF and versioned identifiers. Repeated assay projections are not independent physiological validations. Target ORF2920 maps to P14136-1, without full original clone-sequence proof here. |
| 32814053 | Original NN1.0 and extended rows support 162 target contexts. The tables are gene-level. Historical symbol/accession joins and the two curation conflicts are retained separately from positive-assay evidence. Independent partner records provide class information. |
| 17500595 | Human-brain GFAP peptides associate with an expanded-polyglutamine HTT fragment. Fragment recovery is not a purified full-length wild-type binding demonstration. |
| 27229929, 29892012, 31515488 | Exact supplementary target rows were checked. Summary repetitions are not new experiments; nondisrupted variant entries are not negative wild-type interactions. Historical clone and gene crosswalk limits remain explicit. |
| 33961781, 34819669 | BioPlex/MuSIC AP-MS establishes association in cell-line contexts. Shared datasets and reciprocal table layouts are not independent replication; purified binary contact and current natural-product equivalence are not inferred. |
Four unresolved identities remain explicit: raw54 TP53BP2 Q05BL1 has no established source-accession crosswalk; raw75 names NEK6-4 while the source names NEK6-1; raw105 ZNF19/P17023 conflicts with source GeneID6155/RPL27; raw186 BRK1/Q8WUW1 conflicts with GeneID7428/VHL. Each retains its original identifiers and positive gene-level evidence without silently choosing a different product or gene.
Actual PTHR45652 tree/MSA replay places canonical GFAP at AN268/PTN002509746. Structural activity, organization and filament location derive from AN0/PTN000580114; CMA regulation derives from AN260/PTN002902052. All four IBD records lie on the target path. Target self-evidence is legitimate, and no negative assertion was present in the retained family records. This does not establish universal absence of contrary biology.
Reactome 1253321 maps output1253311 to human P14136 residues1–432 in cytosol. GFAP is the expression target of ERBB4/TAB2/NCOR1 in 1251985, not the transcriptional regulator. The 9613829 CMA pathway is contextual support, with primary mechanistic scope assessed separately.
Raw252 contains one conjunction: part_of enteric nervous system development AND results_in_movement_of migratory enteric neural crest cell. Raw253 contains part_of CMA. All three atoms are preserved as flat schema-supported predicate strings; there are no OR groups. BFO:0000050 is identified by its actual official IRI despite a malformed RO: short identifier in the OLS response.
GO:0005519 is current and nonobsolete in the retained official QuickGO response. GO:0019215 denotes interaction with an assembled intermediate filament, which is not interchangeable with binding an individual cytoskeletal protein. No assembled-filament binding term is substituted.
Each PMID has an explicit full_text_unavailable flag. False means selected authentic original access or an available genuine full cache; it does not mean an exhaustive reading. Papers16625196 and17974005 have available full caches but only their abstracts were assessed. True records abstract-only or no-abstract access. Every reference assessment states its actual scope; three no-abstract clinical entries remain UNVERIFIED.
Five short exact excerpts occur once each in the reference findings. Their original XML-derived abstract bytes, offsets and word budgets are archived. All five native-cache byte offsets are now verified; four excerpts required whitespace-only line-wrap adaptations. Independent partner excerpts are deduplicated by record and field; all other field provenance uses identifiers, evidence arrays and content hashes rather than reproducing complete database text.
The two assembly references have genuine native caches, bringing the source total to 69 while preserving all 67 original source files. PMID6371809 has an abstract-only cache; its scanned original pages supply the inspected full-paper evidence. PMID35511821 has full text, with selected assembly Methods and Results checked. Historical source counts, machine joins and check outcomes remain in the evidence archive and append-only history.
Selected full text for 11567214, 17043438 and 20359319 clarifies mutation and clinical context; the reference assessments state the inspected scope. 23828821 distinguishes PAD2-treated human-brain GFAP from endogenous human-brain peptides. Citrullination identifies GFAP as a modified substrate and does not establish a new catalytic function. No images, raw spectra or supplementary experiments were independently inspected in these native readings.