DLG3/SAP102 is a postsynaptic scaffold. The review retains 456 imported annotation identities and distinguishes a demonstrated interaction from a native cellular mechanism. The three current human products are 817, 512 and 366 amino acids; sequence reconstruction confirms the official alternative-sequence features. The short human products are not the rat SAP102 I1/I2 splice constructs used in neuronal experiments.
The ClinGen Mendelian curation instructions retain supported generic binding when a more informative claim is not established. Different source rows here name different partners: uncertainty about one literal pair does not dispute DLG3 protein binding generally. Actions differ because the measured claims, constructs and results differ, not merely because the evidence codes differ.
Gogl et al., 2022, Methods “Holdup experiments” and “Quantification of binding affinities,” describes recombinant human PDZ domains tested against synthetic peptides. Its Supplementary Data 1 workbook, sheet PDZ-PBM interactome, contains 1,344 measurements involving the three DLG3 domains. The 396 imported partner rows have 1,221 accession-candidate measurement joins: 1,200 match the supplied product sequence using the published natural-ten-mer field, while 21 do not. Three AXIN1 rows additionally match through their actual internal peptide sequence, giving 1,203 sequence-supported scientific selections. Selection does not mean positive binding. Modified motifs and isoform-specific tails remain distinct. The reproducible evidence artifact supplies the workbook extraction, separate assay fields, identity inputs and executable parser. The sheet legend defines COMP as the preferred estimate across assay preparations. LYSC, SAPF and DAPF have different sensitivity thresholds. A zero composite pKd is not a measured zero affinity, and an interaction database’s negative=false flag or literal 1.0x10(molar) encoding does not turn it into a positive measurement.
Positive measurements establish domain–motif recognition at the assayed fragment and modification. They are retained as non-core associations unless independent target-specific context supports an established partner-class function, as for GluN2, FRMPD4/Preso and ERBB4. A blanket peptide-binding replacement describes the reagent rather than that biological role. The measurements do not independently establish full-length physiological interactions, shared cellular compartments, regulation of the partner or functions of all DLG3 products. The 120 unresolved cases include 117 below-quantification pairs and three distinct identity/assay discrepancies. Their uncertainty concerns the literal partner assertion, not paper accessibility. The original Discussion expressly warns that failure to quantify a pair is not biological noninteraction.
Important crosswalk cases are:
Masuko et al., 1999, actual author-posted Methods/Results, tests human DLG3 constructs, bovine brain calmodulin and a rat NR2B peptide. Human host cells, homologous receptor peptides and exact current human protein products are kept separate. Garcia et al., 2000 identifies SAP102 in a human-brain screen against the ERBB4 tail; later PSD-95 co-immunoprecipitation is not attributed to DLG3. The LRP2 and ABCA1 historical originals could not be recovered in full; the former has explicit positive SAP102 evidence in its abstract and curated domain records, while ABCA1 has independent positive native-tail measurements in the 2022 study.
The Preso study, Figure 1 and Results, explicitly includes SAP102 in motif-controlled recognition, pull-down, cellular co-immunoprecipitation and rat-brain complexes. Its independently established spine/F-actin regulatory role supports the cytoskeletal-regulatory partner class; this does not transfer the PSD-95-specific localization experiments to DLG3. The GPR85 study tests receptor-tail constructs. The phosphomimetic peptide-screen study uses human DLG3 PDZ2 and native/Ser-Thr-to-Glu peptide libraries. The mere presence of a phosphomimetic library does not establish physiological phosphorylation dependence for each hit.
The HuRI original, “Generation and characterization of HuRI,” describes sequence-confirmed retesting. All nine imported pairs have three curated assay records, with exact sequence checksums matching 817-aa human DLG3 and the literal partner products. This includes EIF4H-2 (228 aa), DHRS11-2 (181 aa), OSBP2-2 (915 aa), DLGAP2-3 (975 aa) and the 543-aa Q6IS01 product. All nine HuRI pairs are retained as non-core associations. The three records are assay stages within the same study, not three independent physiological replications. Vitronectin’s extracellular context is not reproduced by the yeast assay; EIF4H’s cytoplasmic location is compatible with DLG3, but the native role of that pair remains unresolved. These limits concern functional interpretation, not rejection of the measured binding. Dataset-wide orthogonal sampling is not validation of each individual pair. RPS8 binding is not automatically binding to an assembled ribonucleoprotein complex; the short DHRS11 product is not assumed to retain full-length enzyme activity.
BioPlex, Methods “Affinity Purification of Protein Complexes” and “Identification of High Confidence Interacting Proteins,” reports gene-level AP-MS associations. The exact DLG3–ARHGAP17 and DLG3–ARHGEF26 pairs share interaction accession EBI-54784725 but remain distinct spoke-expanded edges. DLGAP4 is EBI-54638731 and KCNA4 is EBI-54596288. These capture results do not by themselves establish direct partner-class recognition.
The 2025 multimodal-map original links its primary AP-MS graph in Data Availability. Exact edges 609982 and 609986 connect DLG3 bait node 609978 to EVI5 node 572570 and ARHGAP17 node 599600, respectively. The partner nodes have bait=false; the data are not called embedding predictions or pair-specific SEC/imaging validation. ARHGAP17’s multiple Ensembl identifiers remain a product-mapping limit.
Sans et al., 2003, selected actual author-deposited original Methods and Figures 1–3/6–7 text, supports PDZ-mediated receptor-associated trafficking and rat-brain complexes. The single core includes PDZ-mediated receptor-tail recognition as a mechanism of the integrated postsynaptic scaffold, supported also by the controlled receptor-tail assay in PMID:10026200. Sec8 perturbation supports receptor-associated trafficking but does not equal selective human DLG3 knockout. Wei et al., 2015 uses rat neurons and historical 849-residue splice constructs: I2 knockdown raises GluN2A, while GluN2B effects differ between exogenous hippocampal imaging and endogenous cortical biotinylation. These results support contextual receptor positioning, not a universal direction for every receptor/product.
Van Campenhout et al., 2011 and its supplementary Methods support endogenous mouse epithelial Dlg3 biology. The recovered full-length cloning Methods do not explicitly give the tagged construct’s species/accession. PP1 association and Nedd4-dependent trafficking do not make DLG3 a phosphatase or ubiquitin ligase. The separate 2009 PP1-interactome cache remains abstract-only.
The DGKI original explicitly uses mouse SAP102 and rat DGKI constructs, alongside rat-brain complexes. DGKI is a lipid kinase. The corresponding kinase-binding assertion must not be rewritten as protein-kinase binding on that evidence alone. Inherited protein-kinase binding has distinct PAINT grounding.
The tear-proteomics original, Table II, explicitly lists DLG3 among proteins identified from pooled tear samples. This separately recovered publisher table is absent from the normal publication cache despite its unchanged full_text_available flag. Schirmer collection, detergent extraction, gel separation and mass spectrometry support contextual extracellular detection, retained as non-core. The protein list provides no DLG3 peptide sequence or score; no secretion route, extracellular function or contamination claim is inferred. The 1997 NE-dlg study remains abstract/preview-only; expression differences and APC binding alone do not establish the specific proliferation mechanism.
The current PTHR23119 tree places human DLG3 at PTN002508431 beneath every imported IBD node. Actual IBD records and mapped experimental descendants were checked, including rat Dlg3 structural-PSD evidence and mouse Dlg3 receptor-level/polarity evidence. The target’s own experimental grounding and a small donor list are not circularity or weakness. IPR016313 explicitly includes vertebrate DLG1/2/3/4 and matches Q92796 residues 2–817. Its family label does not establish a wrong-paralog transfer. Current ARBA00028600 retains kinase binding but its current condition alternatives do not reconstruct the historical IPR branch; that version limit is recorded without discarding independently supported target function. Rat Q62936 has direct PSD/spine localization evidence supporting the ISS rows.
Typed Reactome paths, read through actual event, complex and member records, resolve Q92796 as DLG3 cytosol entity 6794232 in the DBNL–SHANK context and DLG3 plasma-membrane entity 443609 in NMDA, kainate, RAS-GEF and SALM contexts. Membership does not transfer receptor conductance, GEF activity or actin nucleation to DLG3. The postsynaptic-density-membrane refinement specifies the established synaptic pool and does not exclude other locations.
The machine-fetched publication caches retain their original flags and bytes. Several protected main caches differ from independently fetched originals used for research; neither version was silently replaced. PMID 12738960 is abstract-only in the normal cache despite separately recovered original text. PMID 21920314 and 25555912 have genuine full-text caches. Original recovery for 12713445, 16192269, 19389623 and 9188857 remains incomplete. Exact short quotations appear once at the load-bearing YAML claims; repeated references point back without duplicating quoted text. Quote count is not evidence count. The review proposes no new GO annotations.