EPCAM is a single-pass epithelial surface glycoprotein with a well-supported lateral/basolateral membrane pool. The clearest molecular relationship in the inspected human experiments is association with claudin-7 and stabilization of a claudin-containing pool. Endogenous co-immunoprecipitation, lysosomal inhibition and a binding-impaired transmembrane mutant support this relationship (PMID:23486470, Figures 5–10). Claudin-1 can associate through claudin-7; a ternary recovery does not establish every possible binary contact.
The traditional model in which EpCAM itself forms adhesive trans oligomers is disputed. In the rat-construct experiments, EpCAM without associated claudin-7 increased aggregation and cell-monolayer attachment, whereas association with claudin-7 suppressed those effects (PMID:19276185, Figure 3 and Results pp. 288–292). Later human biochemical and cellular experiments distinguish cis association from trans interactions and fail to detect the proposed homophilic bond under the tested conditions (PMID:30185875, Figures 2–5). The inherited broad adhesion annotation is retained as contextual and non-core. It is not assigned as a process performed by the claudin-binding core. The source guide details species, constructs, controls and access.
Proliferation and differentiation effects depend on context. Human embryonic stem-cell knockdown decreases cell counts and changes lineage output (PMID:19785009), whereas human intestinal cell lines respond differently (PMID:23486470). These are supported regulatory phenotypes; they are not evidence that EPCAM is a cell-cycle enzyme or that a single direction applies in every epithelium. The stem-cell differentiation term is refined because the experiment begins with established stem cells and measures subsequent lineage differentiation, whereas GO:0048863 describes acquisition of stem-cell features.
Released EpICD is associated with FHL2 and a nuclear assembly containing LEF1 and beta-catenin, with transcriptional reporter effects (PMID:19136966, actual main figures and supplement). Human promoter ChIP provides further complex-associated evidence (PMID:20064925). Neither a ChIP signal nor an EMSA from a protein assembly proves direct DNA recognition by EPCAM. Other proteins execute the proteolysis; being the cleavage substrate does not give EPCAM protease activity or participation in the cleavage process. No new process annotation is proposed.
The normal UniProt record and independent official FASTA agree on one displayed 314-residue precursor. The signal peptide is 1–23, processed chain 24–314, extracellular topology segment 24–265, membrane helix 266–288 and cytoplasmic segment 289–314. Sequence SHA256 is e5d527336d82e9309b5ea19a32a85ad410a303db94b245801462e791abd1623c. No alternative splice product is enumerated in the source record. EpEX/EpICD are proteolytic products, and a recombinant cytoplasmic-tail construct does not establish an exact endogenous cleavage boundary.
Human congenital tufting enteropathy variants and altered EPCAM RNA/protein were inspected in PMID:18572020. Its pathogenic exon-skipped product is not added as a normal isoform. The separate Lynch syndrome mechanism described in PMID:19098912 is transcriptional read-through and silencing of neighboring MSH2 after certain EPCAM genomic deletions. That abstract-supported disease mechanism is not attributed to EpCAM protein as methyltransferase or DNA-repair activity.
All 40 imported annotations have manual decisions: 15 ACCEPT, 16 KEEP_AS_NON_CORE, 4 MODIFY and 5 UNDECIDED. Imported terms, evidence, citations, partners and source flags are preserved. Supported generic interactions remain non-core when no narrower function is established. The refinements identify claudin-associated cell adhesion molecule binding, NEK7 kinase binding, plasma membrane, and regulation of cell differentiation. No new annotations or publication quotations were added.
Five uncertainties remain. The ureteric-bud development transfer traces to a mouse expression study whose complete original was unavailable. Three tight-junction annotations retain uncertainty because the older tissue localization experiments were inaccessible and later human intestinal imaging places EPCAM below the tight junction. The fifth uncertainty is the inherited cadherin-binding role in trans-cellular adhesion: the rat donor experiments concern claudin association and attachment, while the independent human fragment-screen hit does not resolve this combined role. Valid PAINT ancestry and the correct UniProt location mapping are preserved; one context's negative localization does not establish a universal absence.
The human interaction screen's actual supplementary rows were inspected for all five partners. These are human cDNA-fragment Y2H results, with no established pair-specific mammalian validation. The CDH1 hit is retained as a non-core screened interaction. Its exact fragment boundaries were not established, so the tested domains are not assumed to be extracellular. The source does not establish cadherin binding across a physiological cell contact. Conversely, the indirect cadherin–cytoskeleton effects in PMID:9382878 do not show that all EPCAM–cadherin binding is impossible. Human T84 experiments in PMID:23486470 reported no tight EPCAM–E-cadherin association (main Results before Figure 2, referring to Supplemental Figure 1B); this is a context-specific limit. Interaction with CDK5R1 does not make its partner the CDK5 catalytic subunit.
The normal fetch produced 40 raw/TSV/seed contexts and 28 publication caches. The genuine Falcon command with Perplexity fallback failed; no provider-named substitute was authored. These notes and the source guide are manual research based on the actual originals and database sources described there. Four additional normal publication caches were genuinely acquired for cited supporting/donor studies; an already present PMID:20064925 was reused. Normal source caches and their acquisition flags were preserved.
Normal validation of the follow-up candidate passed on 2026-10-24, including schema, reference, GOA and ontology-term checks, with four generic-binding advisories for the retained non-core CDH1, CALR, DLST and CDK5R1 interactions. The normal status command reports no mismatch for DRAFT. All 32 normal publication caches and imported source fields remain unchanged.
The core describes claudin association, stabilization and lateral localization; it has no cell-cell-adhesion process limb. PMID:23486470 does more than localize the proteins: Figures 7–10 examine degradation, rescue, redistribution and barrier function. EpCAM depletion lowers total claudin-7/claudin-1 but increases the fraction remaining at tight junctions; claudin depletion suppresses the increased barrier resistance. These experiments support regulation of the claudin pool, not a general claim that the associated complex promotes adhesion. No replacement process annotation is introduced simply to populate the core.
The source guide is a locator index, not independent biological evidence. It remains available for checking the original studies and data, but is no longer cited as support on every annotation. Paraphrased findings now identify the decisive original result and locator on the four principal adhesion/interaction references.
The campaign binding instruction retains supported generic interactions as non-core when a more specific role has not been established, and uses MODIFY when a specific role is supported. The campaign quotation instruction requires preservation of inherited quotations and publication caches without adding new publication quotations. Accordingly, reference-only citations, paraphrased findings and exact original-paper locators are used here. No verbatim-substring validation of a newly added quotation is claimed.
The ureteric-bud transfer and three tight-junction rows remain undecided. An accessible expression-focused abstract does not establish that the complete developmental study performed no functional experiment. The older localization abstract explicitly reports tight-junction staining in human/rat tissues, but its inaccessible images cannot yet be reconciled with the later human intestinal below-junction pool. Neither uncertainty is treated as a demonstrated false annotation.
Regulation of cell differentiation remains the replacement for the source-specific stem-cell differentiation row. PMID:19785009 Figures 4–6 measure altered lineage output from established HES2/HES3 cells while pluripotency and teratoma formation are largely retained. GO:2000036, regulation of stem cell population maintenance, is current, but is not a better description of those measured outputs. PMID:20064925 adds H9 expression/ChIP observations; its functional knockdown is in HCT116 carcinoma cells, not a direct hESC population-maintenance assay.
The released EpICD gene-expression signaling annotation is retained as non-core, matching the nuclear-complex binding assessment. The core remains the membrane claudin-associated function. The source evidence and product boundaries are unchanged.
The full original Results paragraph after Figure 10B/C in PMID:23486470 explicitly discusses residual claudin stabilization by the binding-impaired EpCAMmutHA protein, with residual binding below assay detection and additional stabilization mechanisms both left open. The normal abstract cache is unchanged; the source guide and reference review point to the separately retrieved full original.
A protein-stabilization process annotation is a potential future refinement. No NEW process row is introduced in this follow-up: that would require the separate comparator and ontology-parent checks in the repository guidelines, not simply filling an optional core process slot. The campaign quotation rule is not a blanket prohibition on justified NEW annotations.