2026-05-29

log/2026-05-29 — ad-hoc daily entries

Sub-file of log — see parent for index.

[2026-05-29] ingest | Tyshkovskiy et al. 2026 Nature — universal transcriptomic hallmarks of mammalian ageing & mortality

User-supplied PDF (~/Downloads/s41586-026-10542-3.pdf; doi:10.1038/s41586-026-10542-3; open access). Main text (pp.1–14) + Methods (pp.16–24) read end-to-end. Major Gladyshev-lab paper: >11,000 transcriptomes, >25 tissues, 4 mammals (mouse/rat/macaque/human) → interpretable multi-species transcriptomic clocks (tAge) of chronological age, normalized age, expected mortality (Gompertz log-hazard), maximum lifespan + 23 rodent / 14 multi-species module-specific clocks.

Schema escalation

• R48 — type: biomarker modality: enum gains transcriptomic (first transcriptomic clock page in the wiki). Updated schema-history.md + CLAUDE.md modality line. See schema-history § R48.

Added (4 new pages)

• studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md — verified: true (2026-05-29, claude; main-text + Methods PDF-verified, Extended Data not exhaustively read — scoped). Anchor page.
• biomarkers/transcriptomic-clock-tage.md — verified: true (anchored on the primary PDF). Fills the structural gap: wiki had DNAm + proteomic clocks but no transcriptomic clock. modality: transcriptomic (R48).
• molecules/proteins/gpnmb.md — verified: false + banner (Tyshkovskiy aging claims PDF-verified; general GPNMB biology drafted from training knowledge, queued for verifier). UniProt Q14956 verified via REST.
• molecules/proteins/lgals3.md — verified: false + banner (same scope split). UniProt P17931 verified via REST.

Updated (propagation — 7 existing pages + footnotes)

• molecules/proteins/p21.md — new §“Universal transcriptomic mortality marker (multi-species)”; CDKN1A as top conserved mortality gene + UK Biobank plasma (n=51,276).
• molecules/proteins/igf-1.md — cross-species transcriptomic confirmation Igf1 negatively assoc. max lifespan, declines with age.
• molecules/proteins/klotho.md — new §“Transcriptomic-clock characterization of Klotho-KO progeria (2026)”; metabolic/NRF2-module ageing (not inflammatory); Cdkn1a top driver; cell-type-heterogeneous snRNA-seq.
• interventions/lifestyle/caloric-restriction.md — new para: CR reduces mortality-tAge via metabolic modules (haem/ROS, lipid/PPAR, cholesterol/mTOR).
• hallmarks/chronic-inflammation.md — new §“Inflammation as the dominant mortality module”; immune-module clocks dominate chronic-disease tAge; GPNMB/CDKN1A/LGALS3 plasma mortality markers.
• hypotheses/information-theory-of-aging.md — new §“Tyshkovskiy 2026 — transcriptomic ‘ground zero’ U-curve”; tAge minimum ~E10, rejuvenation phase E8.5–E15; overlap with CR + parabiosis signatures.
• interventions/stem-cell-therapy/in-vivo-partial-reprogramming-therapy.md — new para: reprogramming lowers tAge, strongest in EMT/MET module; transcriptomic vs DNAm clock complementarity.

Gaps surfaced / deferred (→ ROADMAP)

• Secondary universal genes not yet seeded: Cst7, S100a8/9/4/6, Sparc, Nrep, Ddost, Gpx1, Nmrk1, Fmo3, Vcam1, Eda2r, Vsig4.
• New entities referenced but not created: model-organisms/macaca-fascicularis (crab-eating macaque — major data source, no primate model page exists); a heterochronic-parabiosis intervention page (blood-product/); tissues/blood (dangling link from clock page); methods pages WGCNA / scRNA-seq / Mendelian-randomization / elastic-net.
• GPNMB + LGALS3 verified: false → wiki-verifier pass needed (general biology, druggability tier, MR status, GTEx aging-correlation).
• Open scientific gap: transcriptomic tAge intervention-responsiveness in a human RCT untested (the CALERIE-2/DunedinPACE analogue). needs-human-replication

[2026-05-29] ingest | Tyshkovskiy follow-on Batch 1 (4 seeds + propagation)

ROADMAP “From Tyshkovskiy 2026 ingest” Batch 1 — wiki-seeder ×4 (all verified: false, pending verifier):

• added: model-organisms/macaca-fascicularis.md (NCBI tax 9541, AnAge 39yr; gap genome-wide ortholog %)
• added: interventions/blood-product/heterochronic-parabiosis.md (mode: blood-product) + new systemic-milieu-restoration class in intervention-classes + propagated value to plasma-exchange mechanisms
• added: molecules/proteins/cst7.md (cystatin F, O76096, tier 4; fixed spurious dna-damage-response pathway link)
• added: molecules/proteins/eda2r.md (Q9HAV5; corrected anchor cite Barbera 2025 Nat Commun, not 2024)
• propagated: back-links into anchor study page (macaque + parabiosis + cst7/eda2r); wikilinked cst7/eda2r/col1a1/col3a1 on transcriptomic-clock-tage driver list
• new implicit stubs: lysosomal-degradation, conboy-2005-parabiosis-satellite-cells

[2026-05-29] verify | gpnmb.md

Pages verified: 1

• molecules/proteins/gpnmb.md → verified: true (partial scope per verified-scope; 2026-05-29, claude)

Sources checked and status:

• UniProt Q14956 (REST API) — all molecular-biology claims confirmed (572 aa, signal 1–22, ECD 23–498, PKD 240–327, TM 499–519, cyto 520–572, 12 N-glyco sites at confirmed positions, S542-P, cell-membrane/melanosome/early-endosome localisation)
• Rose 2010 PLoS One (doi:10.1371/journal.pone.0012093, PMID 20711474) — ADAM10-as-sheddase + angiogenic ECD migration claims verified via PubMed abstract; archive still pending
• Pahl 2010 Clin J Am Soc Nephrol (doi:10.2215/CJN.03390509, PMID 19833906) — 8.6× monocyte / 17.5× macrophage fold-changes, n=21+22 verified via PubMed abstract; archive still pending
• Saade 2021 Front Immunol (doi:10.3389/fimmu.2021.674739) — local PDF read (8 pages); review claims confirmed; new finding: Saade notes two isoforms (572 aa + 560 aa via alternative splicing) not in original draft
• Chen 2024 Gut (doi:10.1136/gutjnl-2024-331903) — local PDF read (5 pages); FTO/m6A/GPNMB/sEV/SDC4/CD8+ T-cell suppression axis confirmed; n=95 HCC patients
• METRIC trial (Vahdat 2021) (doi:10.1038/s41523-021-00244-6) — Crossref abstract: 327 total, 213 GV : 92 capecitabine (2:1), PFS 2.9 vs 2.8 months — all wiki numbers confirmed; full PDF not available (not_oa)
• Tyshkovskiy 2026 aging/mortality claims — cross-checked against verified studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md (not re-read)
• GenAge human database (build 21) — GPNMB absent; genage-id: null confirmed
• MR literature search (PubMed eutils GPNMB + Mendelian randomization) — found 2 studies:
  • Li 2025 (PMID 40214956, Mol Neurobiol): null MR result for PD risk/progression
  • Guo 2026 (PMID 41851136, NPJ Parkinson’s Dis): MR/SMR/colocalisation supports causal role in PD bone-brain axis

Not verified (closed-access or archive-pending): suda2021 (not_oa), wang2025 (not_oa), colombo2019 (not_oa), chung2014 (archive pending), chung2011 (archive not checked), liang2026 (archive not checked), kim2026 (archive not checked), liu2024 (archive pending), kilavuz2025 (archive pending), yardley2015 (archive pending), zhang2013 (archive pending), yu2020 (archive pending), vdvliet2026 (archive not checked)

Corrections applied (3):

1. mr-causal-evidence: not-tested → partial: Two PD-specific MR studies found (Li 2025 null; Guo 2026 positive for PD bone-brain axis) — neither tests aging/mortality outcome specifically. Field updated in frontmatter + Gaps section + R25 section.
2. 560 aa isoform added: Saade 2021 PDF confirms GPNMB has a 560 aa alternative-splice isoform alongside the canonical 572 aa form. Added to Molecular biology section opener with [^saade2021] citation.
3. Chen 2024 footnote n-detail added: Updated from vague “in-vitro/in-vivo” to include n=95 HCC patient samples + n=6/group in-vivo.

Unverifiable claims (closed-access or pending PDF):

• Suda 2021 (Nature Aging, not_oa): GPNMB as seno-antigen on senescent endothelial cells; senolytic vaccine + lifespan extension in progeroid mice — HIGH-PRIORITY for verification when full-text accessible
• Wang 2025 (Int Immunopharmacol, not_oa): myeloid Gpnmb-KO prevents hepatic LAM formation + MASLD fibrosis reduction
• Chung 2014 (JID, archive pending): DC-HIL ITAM phosphorylation + IFN-γ/iNOS mechanism in MDSCs
• Liu 2024 (Heliyon, archive pending): n=473, serum/urine GPNMB age-correlation + skeletal muscle mass association
• Kilavuz 2025 (OJRD, archive pending): GBA/Gaucher disease plasma GPNMB claims

R25 supersession check:

• No meta-analyses or RCTs (n>100) found that contradict existing wiki framing of GPNMB and aging/mortality
• METRIC 2021 negative trial confirmed (aging-druggability framing stands as tier 2)
• Two new PD-specific MR papers (Li 2025, Guo 2026) noted as “Supersession candidates” for PD-specific context — conflicting results; both consistent with existing wiki framing (neither contradicts the aging-mortality association)
• literature-checked-through: 2026-05-29 set

Downstream propagation needed:

• None required. The mr-causal-evidence correction is self-contained on this page; no other page currently cites GPNMB MR status. The 560 aa isoform addition is cosmetic background.
• When Suda 2021 (seno-antigen / senolytic vaccine) becomes full-text accessible, the [[senolytics]] and [[cellular-senescence]] pages should receive the specific n/outcome details once verified.

[2026-05-29] verify | macaca-fascicularis.md

Pages verified: 1

• model-organisms/macaca-fascicularis.md → verified: true (partial scope per verified-scope)

Corrections applied (3):

1. hESC vs iPSC (body text): wiki body said “human iPSC-derived cardiomyocytes” → corrected to “human embryonic stem cell-derived cardiomyocytes (hESC-CMs)”. Zhang 2023 paper explicitly uses hESCs, not iPSCs. Footnote already had correct term; body was inconsistent.
2. Section heading year: “Zhang et al. 2022/2023” → “Zhang et al. 2023” (PubMed-canonical year is 2023; advance access was Sept 2022 but not the citation year).
3. Menstrual/AMH paper DOI + metadata: footnote had wrong DOI 10.1538/expanim.23-0119 (non-resolving) → corrected to 10.1538/expanim.25-0098 (confirmed via PMID 41423208). Year “2024” corrected to 2026; journal citation updated to Exp Anim 2026;75(2):234-240; author list and sample sizes added from abstract (n=21/22/74 for menarche/menopause/AMH cohorts). Zhang 2023 footnote updated: archive status corrected to “PDF downloaded” (was “download pending”); n detail added (3 young + 3 aged for snRNA-seq; n=8/group phenotypic assays).

Verified against primary sources:

• NCBI taxonomy ID 9541: confirmed (eutils efetch)
• AnAge maximum lifespan 39 yr, LQ 209%, body mass ~6.4 kg: confirmed (genomics.senescence.info)
• T2T-MFA8v1.1 genome size 3,060 Mb: confirmed (NCBI assembly stats GCF_037993035.2)
• Macaca_fascicularis_6.0 size ~2,906 Mb: confirmed (Ensembl REST API)
• Uno 2019 SULT 95-97% + Uno 2020 UGT3A/8A 93-99% identity: DOIs verified via Crossref; titles/journal/year match
• Zhang 2023 Protein & Cell cardiac/FOXP1 claims: full PDF read (PMC OA via PMC10120965)
• Tyshkovskiy 2026 claims: cross-checked against verified studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md (not re-read)

Not verified (closed-access):

• Yang 2024 Cell metformin (not_oa): ~6-yr brain aging regression + Nrf2 mediation claims unverified against full text; abstract confirms title + year + journal; no-fulltext-access
• Sun 2025 Nature Aging cochlear/SLC35F1 (not_oa): claims unverified; no-fulltext-access
• Takebayashi 2026 Exp Anim AMH (not_oa per archive): abstract verified via PubMed efetch

Supersession check: not required for type: model-organism per CLAUDE.md.

Downstream propagation needed:

• None required. The hESC/iPSC error was body-text only on the macaque page (not propagated to any other page). The DOI correction to the menstrual/AMH footnote is self-contained.

[2026-05-29] verify | eda2r.md

Pages verified: 1

• molecules/proteins/eda2r.md → verified: true (partial scope; 2026-05-29, claude)

Sources checked:

• UniProt Q9HAV5 (live API) — confirmed all identity fields
• Barbera 2025 Nat Commun (doi:10.1038/s41467-025-56918-3) — full PDF read end-to-end (10 pages); locally downloaded
• Arif 2025 GeroScience (doi:10.1007/s11357-025-01672-z) — full PDF read end-to-end (20 pages); locally downloaded
• Yan 2000 Science (doi:10.1126/science.290.5491.523) — full PDF read (4 pages); locally downloaded
• Tyshkovskiy 2026 — cross-checked against verified studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md (not re-read)
• Tanikawa 2010 MCR — closed-access (not_oa); not verified; tagged no-fulltext-access
• Bilgic 2023 Nature — download pending; not verified; tagged no-fulltext-access
• Yang 2019 Biosci Rep (PMID 31829409) — confirmed retracted Aug 2024; confirmed absent from page

Corrections made (4 factual):

1. UniProt entry name: TNFRSF27_HUMAN → TNR27_HUMAN (confirmed from live UniProt flat file: “ID TNR27_HUMAN”)
2. Arif 2025 study design: footnote descriptor “meta-analysis of human transcriptomic aging datasets” → “primary transcriptomic analysis (GTEx v8; 40 tissues; young <40 vs aged >65 years) + co-expression network + random-forest ML”. Arif is a primary GTEx analysis, not a meta-analysis.
3. Arif 2025 key finding upgraded in body: “most consistent age-upregulated genes across tissues, overlapping substantially with GTEx-derived aging signatures” → “top-ranked composite cardiometabolic aging predictor (AUC 0.91; #1 above GDF15 in feature-importance ranking; correlated with cholesterol, CRP, body weight, intima-media thickness in SCAPIS-SciLifeLab longitudinal model)”
4. gtex-aging-correlation field: vague “skeletal muscle and blood documented as most robust” → specific: Spearman ρ = 0.372 (rank 1 in pan-tissue multi-tissue analysis; Barbera 2025 Fig. 1b); adipose tissue added as key tissue per Arif 2025

Not verified (partial scope):

• Tanikawa 2010 p53-target claims: closed-access; no-fulltext-access added inline
• Bilgic 2023 NIK-axis claims: download pending; no-fulltext-access added inline

Supersession check (R25):

• PubMed search “EDA2R aging” 2023-2026 returned 16 hits; reviewed top 10 abstracts
• Zhang 2025 (Research; n=24,920; proteomics): EDA2R in top 20 metabolic aging proteins — AGREES with existing framing, not a supersession
• Li 2025 (GeroScience; n=46,047 UK Biobank; proteomics): EDA2R among shared mortality-associated proteins across glycemic strata — AGREES
• No meta-analysis or RCT found for EDA2R as intervention target; no supersession detected
• literature-checked-through: 2026-05-29 already set on page

Downstream propagation needed:

• The Arif 2025 study-design correction (meta-analysis → primary GTEx analysis) affects only this page’s footnote; no other wiki page currently cites [
• The UniProt entry name correction (TNR27_HUMAN) is cosmetic and self-contained

[2026-05-29] verify | Batch 1 close (cst7 + heterochronic-parabiosis)

• molecules/proteins/cst7.md → verified: true (partial). Corrections: MW ~28→~15.5 kDa; cathepsin targets corrected to L/F/K/V (cathepsin C removed as direct target; Daniels 2023 KO confirms); Syage 2024 sign fix (viral control NOT impaired); Daniels 2023 Aβ finding regionalized to subiculum; brensocatib druggability ref removed.
• interventions/blood-product/heterochronic-parabiosis.md → verified: true (partial; Villeda 2014 + Rebo 2016 no-fulltext-access). Key correction: GDF11 direction — Egerman 2015 showed trend toward INCREASE with age (~1.4-fold, p=0.0534), not “does not decline”; added n-values + strain (C57Bl/Ka-Ly5.2) for Conboy 2005; enumerated Villeda 2011 six factors. Now consistent with canonical gdf11 page (which was already correct — no propagation needed).
• All 4 Batch-1 pages now verified: true. GPNMB + LGALS3 remain verified: false (next batch — need background-fleshing before verify).

[2026-05-29] verify | lgals3.md

Pages verified: 1

• molecules/proteins/lgals3.md → verified: true (partial scope; 2026-05-29, claude)

Sources checked and PDF-verified:

• Henry/Lumbers 2022 Circulation (doi:10.1161/CIRCULATIONAHA.121.056663) — full PDF read; local archive. Key MR finding confirmed: Gal-3 positively causal for HF (higher Gal-3 → higher HF risk); 8 proteins total with robust multiverse MR; observational n=3,019 / 732 HF events.
• Henderson 2008 Am J Pathol (doi:10.2353/ajpath.2008.070726) — full PDF read; n=6 per group UUO; Lgals3 KO reduces collagen (p<0.05) and α-SMA (p<0.01 mRNA) confirmed.
• Wu 2022 Cell Mol Life Sci (doi:10.1007/s00018-022-04137-0) — full PDF read; renal fibrosis (NOT cardiac) confirmed; Twist1→Gal-3→M2 polarization confirmed; IgAN human biopsies n=6/group + mouse UUO n=3/group.
• Chalasani 2020 Gastroenterology (doi:10.1053/j.gastro.2019.11.296) — full PDF read; n=162 (54/54/54); HVPG primary endpoint p=1.0 overall; subgroup n=81 p=0.02 (2mg/kg) confirmed.
• Zetterberg 2022 J Med Chem (doi:10.1021/acs.jmedchem.2c00660) — full PDF read; GB0139 Phase IIb NCT03832946 confirmed; GB1211 Phase II NCT05009680 (hepatic impairment) confirmed; wiki-listed NCT05913388 flagged as discrepant.
• Wang 2022 Front Cardiovasc Med (doi:10.3389/fcvm.2022.868749) — full PDF read; IVW OR 1.0002 (p=0.60) confirmed.
• Gou 2023 Front Cardiovasc Med (doi:10.3389/fcvm.2023.1279396) — full PDF read; IVW OR 0.9869 (p=0.8232) for PAD confirmed.
• Pan 2024 Brain Behav (doi:10.1002/brb3.70103) — full PDF read; IVW OR 1.112 (p=0.010) confirmed; simple mode OR 0.910 non-significant — “consistent across methods” corrected to specify which methods were/weren’t significant.
• de Boer 2010 Curr Heart Fail Rep (doi:10.1007/s11897-010-0004-x) — full PDF read; confirmed review paper (not prospective cohort); synthesizes van Kimmenade n=599 AUC 0.72 + Lok n=232 HR 1.95.
• Puigdellívol 2020 Front Cell Neurosci (doi:10.3389/fncel.2020.00162) — first 2 pages read; confirmed review paper; Mer tyrosine kinase opsonization added.
• UniProt P17931 — REST API: CRD end position corrected 250→248; 8 repeats residues 36–109 confirmed; NES 226–241 confirmed.

Not verified (closed-access / OA fetch failed):

• de Boer 2011 (not_oa): n=592 + HFrEF/HFpEF breakdown flagged no-fulltext-access
• Shen 2024 (not_oa): TLR4/NF-κB + JAK2/STAT3 neuroinflammation claims flagged no-fulltext-access
• Xiao 2020 (fetch failed): cardiac TGF-β1/Smad2 bidirectional loop claims flagged no-fulltext-access
• Rahimian 2021 (fetch failed): AD/PD neuroinflammation claims flagged no-fulltext-access
• Srivatsan 2015 (fetch failed): NT-proBNP incremental meta-analysis claims flagged no-fulltext-access
• Tremblay 2021 (fetch failed): pan-phenome MR scope flagged no-fulltext-access

Corrections made (9 substantive):

1. CRD residue range ~118–250 → 118–248 (UniProt confirmed 248)
2. Henry 2022 footnote “n=large” → n=3,019 observational / 732 HF events; MR: SCALLOP n=30,931 + HERMES 47,309 HF cases; running title “Lumbers et al.” documented
3. Pan 2024 “consistent across methods” → qualified: primary IVW + weighted median/mode/MR-PRESSO significant; simple mode OR 0.910 p=0.453 non-significant; specific ORs/CIs added
4. Wang 2022 footnote: IVW OR 1.0002 (p=0.60) added; HF GWAS source (UK Biobank vs HERMES) documented — explains discordance with Henry 2022
5. Gou 2023 footnote: IVW OR 0.9869 (p=0.8232) added; published 4 Jan 2024 (not 2023) corrected
6. de Boer 2010 reclassified from “review/prospective cohort” to review; cohort sizes synthesized (n=599 van Kimmenade, n=232 Lok) documented in footnote and body text
7. Wu 2022 renal vs cardiac correction: wiki incorrectly placed Twist1/Gal-3 axis under cardiac context; corrected to renal; cardiac TGF-β1/Smad claim correctly remains attributed to Xiao 2020 (with closed-access flag)
8. Puigdellívol 2020 body text: Mer tyrosine kinase opsonization added (from abstract); paper correctly characterized as review
9. Henderson 2008 footnote: n=6 per group, strain (FVB/N), key outcomes (p<0.05 collagen, p<0.01 mRNA α-SMA), mechanistic nuance (Lgals3 KO does not affect macrophage recruitment or Smad2/3) enriched

Supersession candidates (R25 — type: protein):

• Cheng W et al. 2024 ESC Heart Fail (doi:10.1002/ehf2.14813; PMID 38698741); meta-analysis of 24 cohort studies; each 1 ng/mL galectin-3 increase → ~4% higher all-cause mortality + ~3% higher combined mortality/rehospitalization. Strengthens but does not supersede de Boer 2010/2011 HF biomarker framing. Added inline to HF biomarker section. Not a contradiction — consistent with existing framing.
• literature-checked-through: 2026-05-29 set.

Downstream propagation needed:

• None identified. The Wu 2022 tissue-context correction is self-contained on lgals3.md; no other pages cite the cardiac claim via Wu 2022.

[2026-05-29] ingest+verify | Tyshkovskiy follow-on Batch 2 (GPNMB/LGALS3 finish + propagation)

• enhanced+verified: molecules/proteins/gpnmb.md → verified: true (partial). Added molecular biology (ADAM10 shedding, PKD domain, 12 N-glyco, 560+572aa isoforms), LAM/DAM cell-type biology, glembatumumab-vedotin/METRIC tier-2 druggability. mr-causal-evidence not-tested→partial (PD MR: Li 2025 null, Guo 2026 positive). Highest-priority unverified: Suda 2021 seno-antigen/senolytic-vaccine (closed-access).
• enhanced+verified: molecules/proteins/lgals3.md → verified: true (partial). Added chimera-type domain architecture, fibrosis mechanism, FDA-cleared BGM Galectin-3 HF biomarker, neuroinflammation, inhibitor pipeline (GB0139/GB1211/belapectin). mr-causal-evidence: partial (Henry 2022 Circulation causal-HF PDF-verified; null PAD/NAFLD MR; +PD). Verifier corrected 9 items (notably Wu 2022 = renal not cardiac fibrosis; de Boer 2010 = review; Puigdellívol Mer-not-TLR4; Henry 2022 n-values).
• propagated: partial-reprogramming §“Transcriptomic-clock readout (Tyshkovskiy 2026)” + hallmarks-of-aging §“Quantifying the hallmarks: transcriptomic module clocks”.
• All three headline universal mortality proteins (CDKN1A/p21, gpnmb, lgals3) now verified.

[2026-05-29] verify | methods/mendelian-randomization.md → verified: true

PDFs read: Davey Smith & Hemani 2014 (HMG R89-R98), Bowden 2015 MR-Egger (IJE 44:512-525), Bowden 2016 I²_GX (IJE 45:1961-1974), Hemani 2018 MR-Base (eLife 7:e34408), Verbanck 2018 MR-PRESSO (Nat Genet 50:693-698), Henry 2022 Circulation (145:1205-1217). Schooling 2025 cross-checked against verified study page. Davey Smith 2003 closed-access (not_oa) — citation confirmed Crossref. STROBE-MR 2021 PDF still downloading (bronze OA) — checklist structure verified via Europe PMC abstract + Crossref.

Corrections (4):

1. STROBE-MR checklist item count: “24-item” → “20 main items + 30 sub-items (50 total)” — abstract-level factual error
2. STROBE-MR footnote citation: “JAMA 328(22):2261-2269” → “JAMA 326(16):1614-1621” — wrong volume/issue/pages (Crossref-verified)
3. Henry 2022 MR footnote example: removed false “co-localisation PP4 >0.8 required” — Henry 2022 used multiverse sensitivity analysis (120 combinations), not coloc PP4; coloc PP4 guidance in body Validation/QC section retained as correctly general best-practice
4. Protein-page count: “144” → “146” (live grep: 146 files with mr-causal-evidence field)

Confirmed correct: I²_GX >0.9 threshold (Bowden 2016 abstract exact language); Verbanck >48% pleiotropy figure (abstract); InSIDE assumption attribution to Bowden 2015; HERMES n=47,309 HF cases / SCALLOP n=30,931 (Henry 2022 Methods); galectin-3 positive direction (causal HF risk-increasing, confirmed Henry 2022 abstract). Schooling 2025 null MR claims verified against already-verified study page (not re-read PDF).

verified: true (partial scope: Davey Smith 2003 closed-access; STROBE-MR 2021 PDF not yet downloaded but abstract-verified)

[2026-05-29] verify | molecules/proteins/nrep.md

Pages verified: 1

• molecules/proteins/nrep.md → verified: true (partial scope; 2026-05-29, claude)

Sources checked:

• UniProt Q16612 (live REST API) — entry name NREP_HUMAN confirmed; gene name NREP confirmed; synonyms C5orf13/P311 confirmed; 68 aa / 7,909 Da / cytoplasmic / Swiss-Prot reviewed; NCBI Gene 9315, HGNC:16834, Ensembl ENSG00000134986.15 all confirmed. No namespace collision; alias PTZ17 (frontmatter) not present in UniProt synonyms but confirmed via Pan 2002 body text.
• Yue 2014 JBC (doi:10.1074/jbc.M114.609495) — full PDF read (8 pages); eIF3b binding via SPR Kd = 1.26 μM confirmed; 5′UTR binding confirmed (not 3′ UTR); IDP character confirmed (PONDR + CD spectroscopy); EBM (eIF3b binding motif) mapped to M-segment aa 23–46 of P311 and eIF3b RRM (F4 fragment aa 161–264).
• Pan 2002 JCI (doi:10.1172/JCI15614) — full PDF read (6 pages); TGF-β1-independent, non-fibrogenic αSMA+ myofibroblast phenotype confirmed in NIH 3T3 and C3H10 T/2 cells; P311 also decreased TGF-β1 production + inhibited TGF-βR2 + decreased collagen 1/3 + decreased MMP-2/9; in-vivo IHC on 7 human wound specimens confirms P311 in myofibroblasts/precursors. Also confirms PTZ17 alias from introduction.
• Badri 2013 JCI (doi:10.1172/JCI69884) — full PDF read (8 pages); hypotension in P311-KO confirmed; telemetry n=8/group; systolic 16.5±3 mmHg + diastolic 14±5 mmHg lower in KO vs WT confirmed (Fig 1A); total + active TGF-β1/2/3 reduced in aorta (ELISA n=15/group) and serum confirmed; P311-transgenic mice develop hypertension confirmed; human hypertensive arteries overexpress P311 (IHC n=18 NT + 21 HT) confirmed.
• Tyshkovskiy 2026 — cross-checked against verified studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md (not re-read): Nrep DOWN-with-age/mortality confirmed; heterochronic parabiosis top-contributor confirmed; embryonic E10 and CR Nrep-specific claim NOT explicitly confirmed in study-page scope — softened to needs-extended-data-verification.
• Liu 2022 retraction (PMID 35154140) — confirmed Retracted Publication via PubMed efetch; retraction in Front Immunol 2026 Feb (PMID 41798934); paper correctly absent from page.
• Fujitani 2004 (not_oa) — closed-access; not verified; no-fulltext-access added to footnote.
• Yao 2016 Stem Cells Dev (not_oa) — closed-access; not verified; no-fulltext-access added to footnote.
• Chen 2022 JID (not_oa) — closed-access; not verified; no-fulltext-access added to footnote.
• Wang 2017 Front Physiol — gold OA; PDF downloaded; abstract-level claims consistent; full read recommended on next pass.

Corrections made (3):

1. Embryonic ground-zero claim softened: “NREP among genes elevated in this maximally youthful transcriptomic state” → qualified with note that the specific Extended Data contribution is not in the verified study-page scope; tagged needs-extended-data-verification.
2. CR claim softened: “CR increases Nrep in liver” → characterised as consistent with the DOWN-with-age designation but magnitude/top-contributor status not confirmed in verified study-page extraction; tagged needs-extended-data-verification.
3. Footnote status updates: Yue 2014, Pan 2002, Badri 2013 footnotes enriched with specific n-values + confirmed claims; Fujitani/Yao/Chen footnotes updated from “Background claim — archive status: not_oa” to “Unverified — no-fulltext-access”; Wang 2017 updated to note PDF available.

Supersession check: protein type — R25 says skip for protein pages. Nonetheless, a targeted PubMed search (NREP/P311/C5orf13, aging/wound healing/fibrosis, 2023-2026) returned 9 hits; none constitute a superseding meta-analysis or RCT contradicting the core claims. New fibrotic contexts (corneal: PMID 42153780 IOVS 2026; renal: PMID 39951942 PRP 2025) are consistent with the existing contradictory-evidence tag for fibrosis/regeneration duality. literature-checked-through: 2026-05-29 set.

Downstream propagation needed:

• None required. The corrected embryonic/CR claims are softened (not contradicted); no other wiki page currently cites nrep.md for these specific details.
• Wang 2017 (Front Physiol) full-PDF read recommended on next pass for the angiogenesis/endothelial-tube-formation claims.

[2026-05-29] verify | methods/single-cell-rna-seq.md → verified: true

Pages verified: 1

• methods/single-cell-rna-seq.md → verified: true (partial scope; 2026-05-29, claude)

Sources checked and status:

• Squair 2021 Nat Commun (doi:10.1038/s41467-021-25960-2) — full PDF read (10 pages, local archive). Pseudobulk superiority claims, false-discovery rate inflation by naive per-cell DE, and the 18-dataset benchmarking all confirmed. Fig. 4e: Wilcoxon finds ~2042 DEGs in null simulation vs pseudobulk edgeR-LRT ~0-50 — supports “thousands of spurious DEGs” language. The “~100–1000-fold” inflation ratio is a characterization of cells:animals ratio (not stated as such in the paper, but derivable; characterization retained with note it is derived, not directly quoted).
• Macosko 2015 Cell (doi:10.1016/j.cell.2015.05.002) — verified via PMC4481139 full text. 44,808 retinal cells confirmed (from 49,300 initial STAMPs; clustered into 39 populations). ~10,000 cells per 12-hour session confirmed as Drop-seq throughput claim. Bead barcode is 12-bp (4^12=16,777,216 sequences); UMI is 8-bp (4^8=65,536 sequences) — wiki says “~3 million unique CBC-UMI combinations” in the 10x workflow description (Zheng 2017), NOT Macosko — no error there since that refers to 10x beads.
• Zheng 2017 Nat Commun (doi:10.1038/ncomms14049) — verified via PMC5241818 full text. ~8–9k cells per channel (up to 8 channels/run). ~68,000 PBMCs from single donor (not “68k+ PBMCs + brain cells”); ~250k total across 29 samples. 14 bp barcode + 10 bp UMI confirmed.
• Tyshkovskiy 2026 — metacell calibration (~100 cells / ~1M reads) cross-checked against verified studies/tyshkovskiy-2026-universal-transcriptomic-hallmarks.md (Methods section read end-to-end at seeding); not re-read from PDF. snRNA-seq of Klotho-KO kidney + brain from one 8-week control + one Klotho-KO mouse confirmed.
• TMS 2020 Nature (doi:10.1038/s41586-020-2496-1) — 23 tissues confirmed via EuropePMC abstract + PMID 32669714. Ages 1, 3, 18, 21, 24, 30 months and ~350k cells / 120+ cell types consistent with sops/finding-singlecell-aging.md (not independently re-extracted from abstract). PDF download pending (green OA) — exact QC thresholds in TMS body text not re-verified in this pass.
• Picelli 2014 Nat Protoc (doi:10.1038/nprot.2014.006) — not_oa; closed-access — not verifiable from local PDF. Abstract confirms “full-length RNA-seq from single cells.”
• Luecken & Theis 2019 Mol Syst Biol (doi:10.15252/msb.20188746) — abstract verified via EuropePMC. PDF download pending (gold OA). Method-review framing confirmed.

Corrections made (3):

1. 10x Chromium throughput: “~10,000 cells/run” → “~8,000–10,000 cells per channel, up to 8 channels simultaneously per run”. The ~10,000 figure actually describes Drop-seq’s stated 12-hour session capacity (Macosko 2015); 10x Zheng 2017 reports 8–9k per channel. Both numbers were being conflated. Added Drop-seq throughput attribution to Macosko 2015 in the same sentence.
2. Smart-seq2 plate format: “384-well plates” / “~384 cells/plate” → “96-well plates (standard implementation; Tabula Muris Senis used 96-well format)” / “~96 cells/plate in standard format; 384-well adaptations exist”. Picelli 2014 is closed-access — cannot confirm from source PDF; 96-well format confirmed as standard via TMS cross-reference. #gap/no-fulltext-access tag added inline.
3. Zheng 2017 footnote: “68,000+ PBMCs + brain cells” → “~68,000 PBMCs (from a single donor across 8 channels) as benchmark dataset; ~250,000 total cells across 29 samples in the study”. The 68k figure is PBMCs only; “brain cells” addition was unsourced in the footnote.

Unverifiable claims (closed-access or pending PDF):

• Picelli 2014 Nat Protoc (not_oa): Smart-seq2 plate format (96 vs 384-well), per-cell sensitivity metrics, full protocol details — tagged no-fulltext-access inline.
• TMS 2020 (green OA, pending download): exact QC thresholds (nUMI >500, nGenes >200, pct_mito <10% for 10x), intronic-read fraction for snRNA-seq component — these numbers are in the Limitations section of the method page and are unverified against TMS Methods text. They are methodologically standard and consistent with field practice, but not confirmed from TMS PDF in this pass.
• Luecken 2019 (gold OA, pending download): body-section claims about normalization/integration best practices verified only at abstract level.
• Tyshkovskiy 2026 metacell threshold (~100 cells / ~1M reads) — sourced from a single paper; not independently verified from PDF in this pass (accepted from verified study page).

Supersession check: type: method — not required per CLAUDE.md (method pages are canonical-data-stable). Squair 2021 checked: 35 papers found in pseudobulk/DE literature 2022–2026; all are application studies adopting pseudobulk, none constitutes a methodological supersession or contradicts the core finding. literature-checked-through: 2026-05-29 confirmed.

Downstream propagation needed:

• The 10x throughput correction and Smart-seq2 plate-format correction are specific to this method page. No other wiki page currently makes the ~10,000-cells/run claim for 10x — no propagation required.
• The sops/finding-singlecell-aging.md SOP references TMS as “~350K cells from 23 tissues … at 1, 3, 18, 21, 24, and 30 months” — consistent with this page’s claims; no change needed.
• If Picelli 2014 full-text becomes accessible: recheck 96 vs 384-well plate format and remove no-fulltext-access tag if 96-well confirmed.

[2026-05-29] ingest+verify | Tyshkovskiy follow-on Batch 3 (3 methods + Sparc + Nrep)

Seed→verify, all 5 now verified: true (partial scope):

• added: methods/mendelian-randomization.md (category: causal-inference; 146 protein pages cite mr-causal-evidence) + pointer in sops/finding-population-evidence.md
• added: methods/wgcna.md (module-discovery method behind the tAge module clocks)
• added: methods/single-cell-rna-seq.md (scRNA + snRNA; metacell→clock bridge)
• added: molecules/proteins/sparc.md (P09486; conserved DOWN regenerative gene)
• added: molecules/proteins/nrep.md (Q16612; conserved DOWN; top parabiosis-rejuvenation driver)
• propagated: clock page links new method pages + sparc/nrep; study page methods + down-gene links
• new stubs surfaced: methods/spatial-transcriptomics, pathways/ecm-remodeling, pathways/vegf-signaling, studies/davey-smith-2003 (→ ROADMAP)
• LESSON: a brief-induced over-claim (Nrep as CR/embryo top contributor) was caught by the verifier — only the parabiosis-top-driver + DOWN-set status are study-page-supported. Brief seeders only with claims actually extracted on the verified source page. feedback-seeder-fabricates-outcomes

[2026-05-29] verify | molecules/proteins/vsig4.md

Verified: true (partial scope — see verified-scope in frontmatter).

PDFs read: Hall 2020 (Aging Cell, doi:10.1111/acel.13219; pages 1–10 read end-to-end); Vogt 2006 (JCI, doi:10.1172/JCI25673; full text read); Helmy 2006 (Cell, doi:10.1016/j.cell.2005.12.039; pages 1–6 read); Tyshkovskiy 2026 (Nature, primary PDF pages 1–9 + verified study page cross-check). Han 2026 (doi:10.3390/life16050732) confirmed via Crossref; abstract verified.

Corrections applied (5):

1. Hall 2020 footnote enriched: added 3.9-fold figure and per-cell 1.9-fold MFI increase; age correlation r values (males r=0.82/0.82, p<0.001/0.002; females r=0.80, p=0.002 for age; PFI r=0.42 p=0.17 not significant); strain noted (C57BL/6J + NIH Swiss).
2. Hall 2020 body: added sex-stratified precision — female PFI correlation (r=0.42, p=0.17) was not significant, correcting implicit over-generalization of frailty correlation to both sexes.
3. Vogt 2006 footnote enriched: specific fold-reductions (2.3-fold CTL, 3.5-fold IFN-γ, ~2-fold IgG), doses (500 µg i.p.), p-values; tissues where expression was absent (intestines, kidney, skeletal muscle, lung, brain, lymph node) per original paper.
4. Helmy 2006 footnote enriched: C3b-binding specificity (not C4b/C1/C2/C3a), ~60% reduction in rosetting in KO, human macrophage populations confirmed.
5. Han 2026: DOI confirmed valid (Crossref 2026-05-29) — removed #gap/unsourced + disclaimer; footnote updated with abstract-level quantitative data (r=0.867, p<0.001 for urinary VSIG4 vs albumin; db/m 38-week surge finding).

Tyshkovskiy claim verified: The paper does explicitly name “Gpnmb, Vsig4, Cdkn1a and Eda2r” as conserved cross-species upregulated genes (page 6, cross-species section). Wiki claim accurate. Clarified in footnote that UK Biobank plasma validation was for GPNMB/CDKN1A/LGALS3 only — not VSIG4.

Complement-system and innate-immune-response pages: referenced in pathways: frontmatter but pages do not exist — stub-creation flagged for lint pass.

Supersession check (R25): PubMed search for VSIG4+aging meta-analysis or RCT (2023–2026) returned 0 results. No supersession candidates found. literature-checked-through: 2026-05-29 confirmed current.

Downstream propagation needed: None required (Hall 2020 numbers were already correctly stated in the body; the frailty-sex correction is a nuance addition, not a sign flip or order-of-magnitude change). The Tyshkovskiy claim on the vsig4.md page is accurate and consistent with the verified study page.

[2026-05-29] verify | molecules/proteins/s100a8-s100a9.md (calprotectin)

PDFs read: Damo 2013 (PNAS), Ma 2017 (IJMM), Gebhardt 2008 (JEM), Bogdanowicz 2024 (Sci Rep), Bonora 2022 (Aging Cell). Tyshkovskiy 2026 claims cross-checked against verified study page (not re-read). Canonical-DB fields confirmed via UniProt REST. Wang 2026 confirmed via PMID 41903862 but not OA; flagged no-fulltext-access.

Corrections (8):

1. complex-subunits: — bare UniProt IDs → wikilinks ["[[s100a8]]", "[[s100a9]]"] per tsc1-tsc2.md precedent
2. Damo 2013 His6 site — His17, His20 of S100A8 → His17, His27 (H27 confirmed in paper Fig 4B)
3. Damo 2013 pathogens — “C. albicans, A. fumigatus” removed (fungi not tested in this paper; Damo 2013 tested bacteria only: S. aureus, S. epidermidis, E. faecalis, P. aeruginosa, A. baumannii, S. flexneri, E. coli); Damo footnote updated accordingly
4. Bogdanowicz 2024 — study design corrected from “cell culture senescence assay + clinical skin study / n=not reported” to three-arm study: ex vivo + 44-woman clinical arm + 30-woman genomic-biopsy arm; footnote fully rewritten
5. Bonora 2022 — n corrected from “~300” to N=35 pre-frail + N=104 frail (=139 total); HR 2.28 (95% CI 1.20–4.31; p=0.012) for all-cause mortality added; MACE borderline-after-adjustment (p=0.067) noted
6. Wang 2026 — #gap/unsourced → #gap/no-fulltext-access (DOI confirmed real via PMID 41903862; not in archive)
7. Pathway wikilink [[nf-kb-pathway]] → [[nf-kb]] (corrected to match actual filename pathways/nf-kb.md in frontmatter and body)
8. Tyshkovskiy 2026 Bullet 3 — companion gene “Ccl5” removed and replaced with “S100a8” (Fig 6g of primary PDF names S100a8, S100a9, Cdkn1a, Lgals3 as shared across all three rejuvenation contexts — Ccl5 not listed); bullet now covers both S100a8 and S100a9 as shared targets

[2026-05-29] ingest+verify | Tyshkovskiy follow-on Batch 4 (calprotectin + Vsig4) — FINAL paper batch

Seed→verify, both verified: true (partial scope):

• added: molecules/proteins/s100a8-s100a9.md (calprotectin complex; S100A8 P05109 + S100A9 P06702; DAMP/TLR4/RAGE; SASP/inflammaging biomarker). Protein-complex page per tsc1-tsc2 precedent.
• added: molecules/proteins/vsig4.md (Q9Y279; CRIg complement receptor; conserved-UP with age; immunosuppressive-yet-rises-with-age tension kept as no-mechanism). Verifier confirmed Hall 2020 adipose-macrophage 13%→52% (3.9-fold) + resolved Han 2026 DOI.
• propagated: clock-page driver line + study related-genes now link both.
• CORRECTION (main agent, re-read PDF p.13-14/Fig 6g): the calprotectin verifier conflated two gene sets. PDF confirms: (a) shared CR+parabiosis rejuvenation = Cdkn1a/Ccl5/S100a9/Lgals3 (S100a9+Ccl5, NOT S100a8); (b) embryonic ground-zero = Cdkn1a/S100a8/S100a9/Lgals3. Fixed calprotectin page bullet 3 + verified-scope. Confirmed all 6 other pages referencing these sets are correct. feedback-abstract-verify-sign-inversion — figure-vs-text conflation by a subagent, caught by primary-PDF re-read.