The provided report is B4GALT1-deep-research-falcon.md (an LLM deep-research synthesis
with non-resolvable internal citation keys, e.g. "wiertelak2025...", "chen2023..."; these
are NOT fetchable PMIDs, so per workflow they are used for orientation only and NOT cited
as supporting_text in the YAML).
Findings USED (concordant with primary literature / UniProt, and independently verified
here):
- Core MF = beta-1,4-GalT transferring Gal from UDP-Gal to terminal GlcNAc, forming
type-2 LacNAc; Mn2+ cofactor; UDP-Gal donor. (Matches UniProt + PMID:16157350.)
- Core localization = trans-Golgi / Golgi membrane, type II topology. (Matches
PMID:6121819, PMID:7744867.)
- Lactose synthase with LALBA is a well-supported, tissue-specific core function.
(Matches PMID:11419947.)
- Downstream cancer migration/invasion, apoptosis, autophagy, inflammatory and
TGF-beta/galectin-8 roles are pleiotropic, context-specific consequences of altered
glycosylation, NOT core B4GALT1 processes, and should not be annotated as dedicated
processes. (Concordant with the "do not over-annotate pleiotropic BP" guidance; none
of these are in the GOA set anyway.)
- The report's framing that cytosol/nucleus localization is unsupported is correct.
Findings REJECTED or DOWNWEIGHTED (and why):
- All specific cancer/disease mechanistic claims (chen2023 integrin alpha6/beta1
substrates in HCC; hsu2024 galectin-8/TGFBR2 in CRC; dai2026 inhibitor apoptosis;
wang2020 glioblastoma apoptosis/autophagy; buduo2021 platelet TPO; tang2023 Alzheimer
glycomics; xu2025 inhibitor Arg187/Glu313) rest on non-resolvable citation keys that I
could not verify against fetchable PMIDs. They are plausible but unverifiable here, are
not in the GOA annotation set, and do not change any annotation action. NOT used as
evidence; recorded here only as context.
- The report repeatedly conflates B4GALT1 with family members (e.g. B4GALT3 in
retinoblastoma) and uses "B4GALT family" effects; I did not transfer any family-level
claim to B4GALT1.
- The active-site residue numbers quoted by Falcon (Arg187/Glu313) differ from the
experimentally verified substrate-binding residues in PMID:2120039 (Tyr284/Tyr309/
Trp310) and PMID:11419947 (His347, Arg359/Phe360/Ile363); I used only the primary,
verified residues and ignored the Falcon residue numbers.
Reviewed all 76 existing GOA assertions, the four inherited cores, both alternative products, the existing Falcon report, and all 21 cited PMID abstracts/metadata and 14 Reactome cache bodies. The source tuples and product identifiers/sequence note are preserved. The proposed review has 46 ACCEPT, 23 KEEP_AS_NON_CORE and 7 MODIFY decisions; it adds no annotation. Eight actions change: rows 9, 12, 16, 34, 42, 64, 71 and 74 (zero-based). Broadness alone is not over-annotation or a reason to classify direct glycan synthesis as non-core.
Identity was independently checked as human B4GALT1, UniProt P15291, HGNC:924 and NCBI Gene 2683. This is distinct from B4GALNT1. The existing normal source records and provider report were preserved; no new fetch or provider call was needed.
The catalytic synthesis now has three substrate/context-specific facets: protein-linked N-glycan galactosylation, LacNAc synthesis from free GlcNAc, and LALBA-dependent lactose synthesis. Manganese dependence is incorporated into enzyme chemistry rather than repeated as an independent binding core. The unchanged UniProt record explicitly states Name=Mn(2+); Xref=ChEBI:CHEBI:29035;. The former manganese core's quote from PMID 16157350 did not itself mention manganese. Its existing IDA manganese-binding annotation remains accepted. Official PDB 2AE7 metadata describes a human M340H catalytic construct with modeled Mn(II); it is not full-length wild-type structural evidence, and an asymmetric-unit/PISA assembly is not proof of a physiological trimer. Coordinates and electron density were not inspected.
The structure cited by PMID:11419947(https://pubmed.ncbi.nlm.nih.gov/11419947/) provides close-ortholog evidence for the lactose-synthase acceptor switch. Official replacement entries 1NKH and 1NF5, superseding 1J8X/1J8W, identify bovine galactosyltransferase and mouse alpha-lactalbumin. That species boundary does not contradict the curated human lactose-synthase assertion: PMID:33805(https://pubmed.ncbi.nlm.nih.gov/33805/) describes soluble human enzyme active on glucose in the presence of alpha-lactalbumin, and PMID:16157350(https://pubmed.ncbi.nlm.nih.gov/16157350/) separately characterizes human enzyme. These cached papers are abstract-only; no complete original protocols are claimed.
Selected full-cache Methods and Results provide direct catalytic support. PMID:27872474(https://pubmed.ncbi.nlm.nih.gov/27872474/) uses human NP_001488.2 and shows enhanced IgG galactosylation in HEK cells. PMID:29133956(https://pubmed.ncbi.nlm.nih.gov/29133956/) explicitly uses human P15291 residues 63–398, with donor/acceptor assays and product analysis, and reports tagged-protein Golgi colocalization in CaCo-2 and COS-7 cells. PMID:38321209(https://pubmed.ncbi.nlm.nih.gov/38321209/) describes a human delta128 GalT construct expressed in E. coli, immobilized and used for glycan/IgG galactosylation. Selected Results of PMID:37632720(https://pubmed.ncbi.nlm.nih.gov/37632720/) support B4GalT1-dependent glycan synthesis; its supplementary construct identity was not independently checked. None of these reads constitutes supplementary-data reanalysis.
Both alternative-initiation products can localize to the Golgi. PMID:1714903(https://pubmed.ncbi.nlm.nih.gov/1714903/) reports relative targeting in the systems examined; it does not establish universal exclusive segregation between a long surface-only product and short Golgi-only product. The raw UniProt entry explicitly records Golgi localization for both. Filopodial localization and several recognition functions are separately marked as transferred from mouse P15535. A localization assertion does not need a dedicated function at that site, so the inherited lack-of-function rationale does not invalidate filopodial or desmosomal CC annotations. The human enterocyte PMID:3917437(https://pubmed.ncbi.nlm.nih.gov/3917437/) cache reports surface/junctional labeling, but its exact desmosome images and full Methods were not inspected; the curator's assignment is retained as non-core.
The broad protein-containing-complex annotation is refined to GO:1902494 catalytic complex, an officially verified child of GO:0032991. The LALBA-associated enzyme is catalytic, and PMID:20378551(https://pubmed.ncbi.nlm.nih.gov/20378551/) supplies additional human-construct association evidence: “GalT, like SiaT, forms Golgi-localized homodimers in live cells.” Its complete available cache contains Introduction, Results and Discussion but omits Experimental Procedures and figure material. Complementation and competition support association; they do not establish one universal endogenous oligomeric state or prove a channeling mechanism. The unverified inherited proposal for a new lactose-synthase-specific GO term is replaced by a question; no new term or complex identifier is manufactured.
The lipid/fibrinogen phenotypes in PMID:34855475(https://pubmed.ncbi.nlm.nih.gov/34855475/) are retained as contextual, non-core assertions with experimental-curator deference. Its abstract links the hypomorph to reduced activity and altered serum glycans, LDL and fibrinogen, but does not resolve specific secretion/clearance pathways. The former blanket claim that B4GALT1 is not a lipid-metabolic enzyme was unjustified: the reviewed record includes glycolipid chemistry. Likewise, a broad membrane-proteomics location is not excessive simply because Golgi localization is separately known. Exact exosome/membrane supplementary protein hits were not independently reverified, and these detections do not show that every extracellular protein molecule is a cleaved soluble form.
Source limitations are recorded in reference assessments. PMID 806951 is a review with no cached abstract; retained enzyme activity is independently established. PMID 2120039, PMID 3917437 and PMID 6121819 repeat their abstract beneath a Full Text heading rather than supply complete Methods. PMID 10900002 explicitly uses B4GALT1 as a Golgi marker despite its FUT6-focused title, but its cached body contains only Abstract and Discussion. Reactome 2025723/2046265/2046298/975919 repeat an overly exclusive isoform-localization statement, and Reactome 5653878 has an apparent product-name typo in its summary; neighboring events and primary biochemistry establish lactose as the product. Raw caches remain unchanged.
Correction to the earlier Falcon integration notes: the preserved report includes a DOI-based bibliography that resolves its internal citation keys. It was too broad to dismiss all keys as unresolvable. A different set of residues in one mutagenesis study also does not by itself contradict other proposed structural residues. The provider report remains an orientation source; specific mechanisms require primary evidence, and family/paralog examples are not automatically transferable to B4GALT1.
All 76 original annotation objects, their evidence codes and reference identifiers are retained, with 46 ACCEPT, 23 KEEP_AS_NON_CORE and 7 MODIFY decisions. The three catalytic cores remain glycopeptide galactosylation, free-GlcNAc N-acetyllactosamine synthesis, and alpha-lactalbumin-dependent lactose synthesis. There is no lactosylceramide core. The long and short initiation-product identifiers and sequence distinction are unchanged; their descriptions now focus on biology. Mouse-derived filopodial/surface-recognition details and the limits of the F9 targeting evidence remain qualified in this journal and the relevant reviews.
The carbohydrate-metabolism IEA now refines to GO:0009312 oligosaccharide biosynthesis, an actual descendant in the checked GO 2026-03-25 release. GO:0006487 is a valid biological role but is not a subclass refinement of GO:0005975 in that release. Formation of the disaccharides LacNAc and lactose supplies a positive chemical basis for retaining the existing GO:0009312 NAS row. The broad NAS molecular function is refined to all three already represented catalytic activities, with their different acceptors and alpha-lactalbumin requirement made explicit. This preserves source provenance and adds no NEW annotation.
The free-GlcNAc and lactose assertions attributed to PMID:16157350 remain accepted on curator attribution plus independent human biochemistry in PMID:33805. The inspected 2005 abstract directly describes branched N-glycan acceptors; it is not represented as exposing every free-GlcNAc or lactose assay. Complete short evidence clauses are centralized in reference findings, while annotation and core citations preserve the source links without repeatedly clipping the same quotation. The 11 joined index comments and 24 trailing-whitespace lines are repaired; all 76 annotation markers are on separate lines.
The fibrinogen and LDL IMP rows consistently describe non-core physiological associations. The reduced enzyme activity and glycan changes do not, from the inspected abstract alone, establish a specific secretion/clearance mechanism. Manganese binding remains accepted as part of catalytic function, with the human M340H soluble-construct and metal-evidence boundaries retained. Five proteomics references are identifier/GOA-attribution verified; their exact supplementary B4GALT1 hits and cleavage states remain uninspected.
A bounded proposed lactose synthase complex term replaces the previous question, superseding the earlier journal decision to withdraw the proposal. Its parent is the verified catalytic-complex term. The candidate distinguishes human biochemistry from bovine/mouse structural evidence and does not imply that alpha-lactalbumin is required for all B4GALT1 reactions. Current-ontology duplicate checking remains necessary before an ontology submission. Raw publications, Reactome entries, UniProt, GOA and the existing Falcon report remain unchanged.
The 76 source assertions and all decisions remain unchanged (46 ACCEPT, 23 KEEP_AS_NON_CORE, 7 MODIFY), as do the two products and three catalytic cores. This entry supersedes the earlier decision to centralize shared excerpts in reference findings. Meaningful evidence excerpts now accompany all three cores and the kinetic-preference, initiation-product targeting, substrate-binding and subcellular-localization reviews that most need them. Repeated finding-level excerpts and their redundant summaries are removed; reference identifiers, titles, scientific review judgments and contextual reading notes are retained. Shared sources remain cited at the other relevant annotations without duplicating their excerpts.
PMID:16157350 now anchors the wild-type substrate-preference comparison directly at its glycopeptide-activity row. PMID:27872474 supplies an additional core anchor from the already inspected IgG glycoengineering Results, with human B4GALT1 construct identity verified in the selected Methods. PMID:33805 retains free-GlcNAc, ovalbumin and alpha-lactalbumin-dependent glucose acceptor context. The ortholog structural boundary for PMID:11419947 remains explicit in the reference review. The surface-targeting and enterocyte-localization excerpts retain their experimental context; they do not assert exclusive isoform compartments or an established adhesion mechanism.
The entire available abstract of PMID:2120039 was read, but it is explicitly truncated at 250 words; the reference reading note now states that limit. The other newly anchored early papers were read at complete cached-abstract level, while PMID:27872474 was read in selected Results and expression Methods. No additional full-paper, image or supplementary-data reading is claimed. The quotation audit counts every occurrence in the proposed YAML and adds no quotations in this journal entry; each source remains within 25 quoted words. Previously authored journal bytes are preserved.
The current PR body already identifies glycopeptide galactosylation, free-GlcNAc N-acetyllactosamine synthesis and lactose synthesis. Its reported lactosylceramide wording is stale reviewer feedback, so no biological change is made on that basis. No raw UniProt, GOA, publication, Reactome or provider file is changed.
Restored ten finding summaries across nine references after checking the available primary evidence. This supersedes the previous entry's removal of those summaries: a finding statement records the interpreted result and remains useful when its excerpt is attached to an annotation or core. The restored findings are paraphrases; the existing meaningful excerpts remain in their reviewed locations without duplication.
PMID:33805 now again records the human soluble-enzyme assays with free GlcNAc, glycoprotein acceptors and alpha-lactalbumin-dependent glucose, together with the fluid-isolation result. Its three IDA catalytic rows already carry explicit source links. Row 23 additionally links PMID:33805 for the independent human corroboration stated in its review; the ortholog boundary of PMID:11419947 remains explicit. Row 46 now links PMID:16157350 directly for N-glycan maturation. The other PMID:16157350 activity rows retain their original attribution and separate human corroboration, without claiming that unread free-GlcNAc or lactose assays were inspected.
The restored summaries preserve source-specific limits: relative initiation-product targeting, the truncated PMID:2120039 abstract, localization without a demonstrated adhesion function, and tubulin association without proof of a universal retention mechanism. Eight available abstract records were read; PMID:2120039 is truncated and three caches repeat abstract text under a Full Text heading. For PMID:27872474, the relevant galactosylation Results and human-expression Methods were checked. No additional complete-paper, image or supplementary-data access is claimed.
All 76 original source assertions, their decisions, two products and three cores are unchanged. Existing journal bytes are retained. No new quotations are added; the aggregate quotation inventory remains within 25 words per source. The current published PR body already identifies glycopeptide galactosylation, free-GlcNAc N-acetyllactosamine synthesis and lactose synthesis. No raw input, provider report or cached source is edited.
Short excerpts from the existing source abstracts now accompany the five requested annotations in the B4GALT1 review. This changes where the evidence is displayed; all 76 source assertions, their actions, two products, reference findings and three core biological claims remain unchanged.
| Annotation row | Existing source | Evidence now displayed at the annotation |
|---|---|---|
| 23, lactose synthase activity | PMID:11419947(https://pubmed.ncbi.nlm.nih.gov/11419947/) | Alpha-lactalbumin promotes glucose binding to the catalytic component. The separately established bovine/mouse structural boundary remains explicit; the abstract does not identify those species. |
| 46, protein N-linked glycosylation | PMID:16157350(https://pubmed.ncbi.nlm.nih.gov/16157350/) | Galactose transfer to branched N-linked glycoprotein chains. Row 45 keeps a shorter excerpt identifying the wild-type kinetic preference; the distinction from M340H structures remains in the review. |
| 67, glycopeptide galactosyltransferase activity | PMID:33805(https://pubmed.ncbi.nlm.nih.gov/33805/) | Ovalbumin and desialylated ovine mucin as acceptor substrates. |
| 68, N-acetyllactosamine synthase activity | PMID:33805 | Activity with free GlcNAc. |
| 69, lactose synthase activity | PMID:33805 | Glucose in the presence of alpha-lactalbumin. |
The second core's PMID:33805 excerpt is redistributed across rows 67–69, where each acceptor directly accompanies its annotation. The free-GlcNAc core retains PMID:33805 and PMID:2120039 as evidence; row 68 displays the relevant acceptor excerpt, and row 63 retains the recombinant-human-enzyme substrate-binding excerpt. No donor, product or reaction claim changes. The third core retains a direct PMID:11419947 excerpt for the 1:1 lactose-synthase assembly, while row 23 displays the glucose-binding mechanism and row 69 the independent human-enzyme assay context. All original core reference links remain present.
The complete available cached abstracts of PMID:33805, PMID:11419947 and PMID:16157350 were reread for this redistribution. The available PMID:2120039 abstract was also reread and remains explicitly truncated. No complete article, new structure, figure image or supplementary dataset was inspected. Each omission marker joins ordered source spans without supplying missing words. The proposed YAML quotation inventory counts repetitions and remains within 25 words per PMID under both whitespace and punctuation-splitting counts; this new journal entry adds no quotations. Earlier journal text is preserved as historical material and is not included in that claim about the proposed YAML. All cached source bytes and availability flags remain unchanged.
The current review feedback identifies evidence-presentation gaps while retaining the biological judgments. The proposed follow-up keeps all 76 actions, source annotation objects and three core-function claims. Row 45 now states only the named tenfold Km comparison, and its excerpt names both comparator arms. Row 68 restores the assay sentence subject. Row 69 keeps its human soluble-enzyme excerpt and adds the immutable UniProt lactose reaction as separate corroboration. No human assay is inferred from that curated record alone.
Core 2 retains both original PMID links and gains the immutable UniProt free-GlcNAc reaction. Core 3 retains its 1:1-complex anchor and both original PMID links, with an additional UniProt mechanism excerpt for the LALBA-dependent specificity change and high-affinity glucose binding. The reaction excerpt keeps the source record's line continuation verbatim; it does not change any chemical identifiers. These are curated-record corroboration of existing claims, not new experiments or annotation decisions.
The existing good rows 23/46/67/69 anchors and all reference findings remain intact. PMID 2120039 is still limited to its explicitly truncated abstract, also repeated below the cache's Full Text heading; no full-paper or direct product-assay reading is claimed. Existing source-use limits are retained. The complete proposed YAML quotation inventory counts repeated occurrences, including the UniProt cofactor excerpts already present, and remains at most 25 whitespace-delimited quoted words per PMID and for the UniProt record. The unchanged generated deep-research quotation is tracked separately. Source-order and whitespace-normalized verbatim matching were checked independently for every retained and added excerpt. Chemical strings are reported as source text, not prose word fragments. Historical notes stay unchanged.
The complete cached abstract of PMID 3917437 was reread for the three localization claims. It reports human duodenal biopsy immunolabeling and quantitative orientation toward the external plasma-membrane surface, lateral membrane labeling, and intense brush-border labeling. Rows 70 and 73 now have brief source-specific anchors. Row 71 retains a shorter excerpt linking the lateral plasma membrane to labeled interdigitations. The three excerpts use 10, 7 and 8 quoted words respectively, totaling 25 words from this source across the proposed YAML. All three localization decisions remain KEEP_AS_NON_CORE. The cache records abstract-only extraction and repeats the abstract below its Full Text heading; no complete-paper, figure or supplement inspection is claimed.
The core 2 curated UniProt reaction excerpt now uses an explicit ordered omission to skip the flat-file CC continuation prefix while retaining both parts of the chemical name. The chemistry and core-function claim are unchanged. Row 75 now calls oligosaccharide biosynthesis an accepted annotation without describing it as an additional standalone core-function entry. The historical lactose-synthase-complex proposal regains its skos:relatedMatch link to RHEA:12404 after checking the matching glucose-to-lactose reaction and Rhea cross-reference in the immutable UniProt record. This relates the complex proposal to its catalyzed reaction; it does not equate a molecular complex with a reaction or create a new GO annotation.
All 76 machine-sourced annotation objects and their 46 ACCEPT, 23 KEEP_AS_NON_CORE and 7 MODIFY actions are preserved, as are the three core claims, original references, prior good anchors and historical notes. There remains one proposed term and no NEW annotation row. No new enzyme classification or scaffold claim is introduced. PMID 2120039 remains limited to its truncated cached abstract, repeated under Full Text. The standing source-use limit is retained: repeated supporting-text occurrences are counted, and every PMID and the UniProt record remain within 25 quoted words. All excerpts pass ordered, whitespace-normalized source matching.
Row 71 now states the lateral-membrane/interdigitation localization directly supported by its retained excerpt. The broader junctional-region and basal-gradient details are no longer asserted in that row summary. This narrows the presentation without changing the basolateral localization decision. Rows 70 and 73 and their excerpts remain intact. The PMID 3917437 cache is abstract-only despite a repeated abstract below the Full Text heading; no complete-paper or image inspection is claimed.
For row 29, selected Results and Methods of the already cached full text of PMID 38321209 were read. The authors used a human GalT catalytic construct with an N-terminal deletion of 128 residues, expressed as an MBP/AviTag fusion and immobilized after biotinylation. The Results describe a free-GlcNAc activity reaction confirmed by MALDI-TOF MS/MS. The construct paragraph and the immobilized-enzyme activity method were also inspected. The latter repeats GlcNAc at two concentrations, so no concentration-specific inference is made. No supplementary sequences, raw spectra or images were inspected. This provides an additional experimental anchor for the accepted catalytic activity without claiming a new reading of the complete PMID 16157350 paper. Its original EXP attribution is retained, and its cached abstract remains bounded to the N-glycan acceptors it actually describes. PMID 33805 remains independent human-enzyme corroboration.
Only the summaries of rows 29 and 71 and the additional support entry on row 29 change. Every prior excerpt, all 76 source annotation objects and actions, the three core functions, two alternative products, reference findings and the one proposed term are preserved. Quotation totals remain within the existing per-source limit; no other anchor is shortened or displaced. No source file was changed or newly fetched.