Open questions
What we do not know yet: questions research has not settled, studies that contradict each other, and the points in our own notes still marked “to verify”.
Research questions 76
- Are somatic mutations numerous enough to explain the functional decline of normal aging, or are they mainly a cancer risk?
-
What matters most: the persistent DNA lesion (alarm signal, senescence) or the fixed mutation?
Schumacher 2021 leans towards lesions.
- Do mitochondrial DNA mutations reach a threshold that impairs function in normal human aging?
- Do mutated clones (CHIP, loss of Y, skin or oesophagus clones) cause disease or are they only markers? Should they be treated?
- How can DNA repair be improved without helping cancer cells survive?
- How can the DNA damage load be measured routinely (single-molecule sequencing, loss of Y, CHIP)?
- What matters: average telomere length measured in blood, or persistent telomeric lesions in tissues?
- Is white blood cell telomere length a good biomarker, despite large variation between people and between measurement methods?
- Can telomerase be activated transiently or in a targeted way without raising cancer risk?
- Does telomere shortening contribute to normal human aging outside telomere syndromes, or only in some organs (blood, lung)?
- Do smoking, stress and obesity really shorten telomeres, or are these statistical confounders?
-
Are the loss of heterochromatin around telomeres and the age-related methylation changes in subtelomeric regions confirmed in humans?
Marked [to verify] in file 02, no source.
-
Is there a link between telomere wear and dysbiosis?
Marked [to verify] in file 02: link not established, no source.
-
Do epigenetic clocks measure a process that causes aging, or just a reflection of it?
Yang 2023 suggests causation in mice; not shown in humans.
- Which clock should serve as a trial endpoint, given that they disagree (CALERIE: effect on DunedinPACE, not on GrimAge)?
- How can reprogramming be made safe (dose, duration, tissue, vector) without teratomas or loss of cell identity?
- Where is the "youth information" that reprogramming restores stored?
- How much of epigenetic drift is due to chance, and how much to a program?
-
Do microbiome metabolites (butyrate, a histone deacetylase inhibitor) influence the epigenome during aging?
Marked [to verify] in file 03: link poorly documented, no source.
-
Do NAD+ precursors have an effect on epigenetic clocks in humans?
Marked [to verify] in file 03: no effect shown, no source.
-
Are large aggregates a cause of toxicity, or a defence that "stores away" more toxic forms (oligomers)?
Removing plaques gives only a modest benefit in Alzheimer.
- Can HSF-1, the proteasome or CMA be boosted in humans without helping cancers, which also rely on these networks?
- How can proteostasis be measured in a living person, beyond specific imaging (amyloid PET) or blood markers (p-tau)?
- Does the causal evidence on lifespan, mostly from the C. elegans worm, carry over to humans?
- Should treatment start before symptoms, as the anti-amyloid trials suggest?
- How can autophagic flux be measured in humans (blood, biopsy) so that an intervention can be dosed?
- Can selective autophagy (mitophagy in muscle, aggregates in neurons) be switched on instead of a global effect?
- Switching on autophagy can help an existing tumour survive: in which patients is that acceptable?
- Which rapamycin dose and schedule targets mTORC1 in humans without the mTORC2-related side effects (high blood sugar, immunosuppression)?
- If the lysosome is saturated, stimulating autophagy is useless: is TFEB a better target?
-
Should rapamycin be taken intermittently in humans to spare mTORC2, and which biomarker can show mTORC1 inhibition in tissues?
Selective mTORC1 inhibitors still need clinical validation.
- Is the anti-aging effect of metformin real, or an artefact of observational studies (survivor and comparator bias)?
-
Will the TAME trial of metformin ever start?
Marked [to verify] in file 06: no primary source; no ClinicalTrials.gov record on 2026-10-19.
- Does low IGF-1 protect or weaken older people? The effect seems to depend on age, sex and tissue.
- Why do many interventions (acarbose, 17alpha-estradiol, S6K1, SIRT6) act so differently in males and females?
- Caloric restriction, intermittent fasting or protein composition: what really matters, given that long-term adherence is a major obstacle?
- Sirtuin activators (resveratrol, STACs) did not deliver: are sirtuins still a valid target?
- Is NAD+ really limiting in humans? Precursors raise blood NAD+ but functional gains are thin or absent; which tissues benefit?
- In humans, is mitochondrial decline intrinsic aging or mostly the result of lower physical activity?
- Must a threshold of faulty cells be reached before a tissue suffers, and how can mtDNA mosaicism be measured in a living person?
- Which ROS should be reduced, where and when? Targeted antioxidants (MitoQ, SkQ1) still need solid clinical endpoints.
- Does urolithin A really act through mitophagy in humans? Direct tissue measurements are missing.
- What standard, accessible measure of mitochondrial health could be used (phosphorus spectroscopy, biopsy respirometry, blood mtDNA)?
-
Elamipretide received FDA accelerated approval (19 Sept 2025) for Barth syndrome, the first mitochondria-targeting drug: will it show benefit in aging-related conditions?
Source is an FDA press release (no DOI), so no claim was created; confirmatory trial required.
- Which marker or validated blood biomarker can identify senescent cells across tissues, to select patients and track effects? p16, p21, beta-galactosidase and SenMayo overlap poorly.
-
Why are human senolytic results disappointing so far (UBX0101 failed, D+Q modest and subgroup-limited): dose, tissue, patient selection or concept?
UBX0101 phase 2 failure (Aug 2020, 183 patients, knee pain at 12 weeks) is sourced by a Unity Biotechnology press release (no DOI).
- Senescent cells differ between tissues and have different weaknesses: can any senolytic work broadly?
- What are the long-term effects of removing senescent cells, given their useful roles (wound healing)?
- Is it better to kill senescent cells or silence their SASP, and for which disease?
- What are the cost, safety and durability of senolytic immunotherapies (CAR-T, vaccines) in humans?
- How can reprogramming be made safe in humans over years (no teratomas, no loss of identity)? Which dose, tissues and delivery?
- Is stem cell exhaustion a cause or a consequence of tissue aging, and how much is intrinsic versus niche in humans?
- Which simple biomarkers of stem cell health exist in humans, beyond CHIP sequencing?
- Stimulating stem cells can favour mutated clones (CHIP, cancers): how to rejuvenate without selecting bad clones?
- Do mouse results (often progeroid) carry over to slowly renewing human tissues?
- Which blood factors really matter? Hundreds of proteins change with age; few are validated and some (GDF11) are disputed.
- Is removing harmful factors from old blood more effective than adding young ones in humans, as mouse data suggest?
- Does a plasma exchange have a lasting effect, or must it be repeated indefinitely?
- Can extracellular matrix stiffening be reversed (cross-link breakers)? Few effective tools exist.
- Human plasma trials (AMBAR, PLASMA) target diseases, not aging itself: which validated outcomes could measure aging?
- Which target (IL-1beta, IL-6, NLRP3, STING...) reduces inflammaging without weakening defence against infections and cancers?
- Would CHIP carriers (especially TET2) benefit most from anti-IL-1beta/IL-6 drugs? A hypothesis from secondary analyses, to test prospectively.
- Is inflammation a driver of aging or mainly a marker? Randomized trials on aging outcomes are missing.
- Can iAge and other inflammatory clocks be validated as trial outcomes?
- Can the expansion of mutated blood clones (CHIP) be prevented? No validated treatment so far.
- In humans, is dysbiosis a driver of aging or a reflection of disease, drugs and diet?
- What is a "good" microbiome in old age? It varies by country, diet and person; no consensus definition.
- Are FMT safety (pathogen transmission) and durability established in older humans outside infection?
- Do probiotics affect aging? Most commercial strains have no proven effect; effects depend on strain and person.
- Which metabolites (bile acids, butyrate, urolithin A, TMAO) carry most of the microbiome effect?
-
Strehler criteria (aging is universal, intrinsic, progressive, harmful) vs disease: where is the boundary?
[to verify] Strehler 1962, Time, Cells and Aging: book without DOI, not verifiable online; not ingested as a source.
-
Mutation accumulation (Medawar): does additive genetic variance in mortality rise with age, as predicted?
[to verify] Medawar 1952, An Unsolved Problem of Biology: lecture/book without DOI; not ingested as a source.
- Is there a single root cause of aging (damage, programs, information loss) or is it an emergent property of a network?
- Different clocks applied to the same person can disagree: which dimensions of aging does each capture?
- Training cohorts for aging clocks are mostly European: do they hold in other populations?
- How can one prove that a treatment slows aging in humans without waiting decades for deaths? No biomarker is yet validated for this.
Contradictions 16
- In mice, lengthening telomeres does not increase cancer; in humans, genetically longer telomeres raise the risk of several cancers. The text notes that mouse results do not transfer directly.
- Twins suggest a drift driven by chance and environment; the Horvath & Raj theory sees a developmental program that keeps running. The text asks how much each contributes.
- One partial reprogramming protocol gave no tumours, while full or poorly controlled reprogramming gives teratomas and cancers: safety depends on dose, duration and tissue.
- Spermidine extends the life of several animals and is associated with lower mortality, but the SmartAge randomized trial finds no effect on memory in humans.
- TORC1 inhibition reduced infections in phase 2a/2b, but the RTB101 phase 3 missed its primary endpoint.
- Observational data suggest metformin may lower mortality, but in a randomized trial it blunts the benefits of exercise.
- A mitochondria-targeted antioxidant improved vascular function in a small trial, while classic antioxidants give no benefit and slightly raise mortality in a meta-analysis of trials.
-
D+Q improves function and survival in old mice, but in a randomized human trial it had no overall effect on bone (benefit only in a high-burden subgroup); UBX0101 also failed in knee osteoarthritis (press release, no DOI).
Transplanting a few senescent cells into a young mouse is enough to cause loss of physical function; in old mice D+Q improves physical function and extends remaining survival by about 36%. ≠ In a randomized phase 2 trial (60 postmenopausal women), D+Q had no overall effect on bone, with a modest benefit in women with a high senescent-cell burden.
- Removing senescent cells extends mouse life, yet transient senescence is needed for wound healing: removal may impair useful functions.
- Continuous OSKM causes teratomas, while a long-term partial protocol in normal mice found no tumours: safety depends on the protocol.
- GDF11 was presented as a young rejuvenating blood factor, but a contradictory study found it rises with age and inhibits muscle regeneration.
- Young plasma improves memory in old mice, but the small PLASMA human trial found no proven efficacy.
-
CANTOS shows that lowering inflammation alone reduces age-related disease in humans, but it raises fatal infections: dampening inflammation weakens an already ageing immune defence.
With age the body keeps up a low but permanent inflammation without infection or injury ("inflammaging"), while the immune system becomes less effective (immunosenescence). ≠ In the CANTOS phase III trial, canakinumab (anti-IL-1beta) cut cardiovascular events by 15% in post-heart-attack patients with high CRP, but increased fatal infections.
- Urolithin A activates mitophagy and helps animals, but in older humans the primary walking outcome was not met (only endurance improved).
- The 12 hallmarks are used as the reference taxonomy, while a published critique argues they lack causal hierarchy and overweight damage.
-
Damage theories (free radicals) see aging as accumulated molecular damage; hyperfunction theory sees it as growth programs running on, not mainly damage.
Harman proposed that aging comes from cumulative damage by free radicals from oxidative metabolism; antioxidant trials in humans did not extend life. ≠ Blagosklonny proposes that aging is driven mainly by growth programs (notably mTOR) that keep running after development, causing cellular hyperfunction; the theory predicted that rapamycin would slow aging.
Points “to verify” in our notes 82
- **Unverified points, marked [to verify] in the files:**
- | 3. Epigenetic alterations | Telomeres are surrounded by heterochromatin; its loss with age weakens their protection. Subtelomeric regions change their methylation [to verify] (for humans). |
- | 12. Dysbiosis | Link not established [to verify]. |
- | 12. Dysbiosis | The microbiota produces metabolites (butyrate, a histone deacetylase inhibitor); link with aging poorly documented [to verify]. |
- | NAD+ precursors | Sirtuins | Preclinical; human phase 1-2 trials | See file 01; no demonstrated effect on clocks in humans [to verify] | — |
- | Metformin | AMPK / mitochondrial complex I | Approved (diabetes); TAME trial designed but not launched: no registration on ClinicalTrials.gov (search of 19/10/2026), AFAR page offline; several 2026 secondary sources say it is still unfunded [to verify]: no primary source | Diab
- - **Classic distinction**: aging is universal, intrinsic, progressive and deleterious (Strehler's criteria); a disease affects a subset of individuals [to verify — Strehler 1962, *Time, Cells and Aging*, no DOI]. This boundary fades once one accepts that age-related diseases are
- | **Mutation accumulation** | Medawar (1952) [to verify — *An Unsolved Problem of Biology*, lecture, no DOI] | Deleterious alleles with late effects accumulate by drift, because selection does not eliminate them | Additive genetic variance of mortality increasing with age |
- **Sources marked [to verify]** (books without DOI, not verifiable online): Strehler 1962 (*Time, Cells and Aging*), Medawar 1952 (*An Unsolved Problem of Biology*).
- Background detail and the full biology are in [`literature/aging-map/`](../aging-map/00-overview.md). This brief does not repeat it. Every key claim cites a source whose DOI was checked on Crossref, or an official link. Anything not confirmed by a primary source is marked **[to v
- | 7 | Metformin | Metformin | UKPDS 34: in overweight T2D, lower all-cause mortality vs conventional therapy ([UKPDS 1998](https://doi.org/10.1016/S0140-6736(98)07037-8)). Observational: diabetics on metformin survived as long as matched non-diabetics ([Bannister 2014](https://do
- | 16 | Combinations | e.g. rapamycin + acarbose, exercise + drug | No completed human aging trial of combinations found **[to verify]** | Rapamycin + acarbose (ITP): median lifespan +28 % females, +34 % males ([Strong 2022](https://doi.org/10.1111/acel.13724)). Trametinib + rapam
- 1. **Aging is not a disease indication.** Regulators approve drugs for diseases. TAME was designed to work around this by using as its endpoint the time to the first of several age-related diseases (heart disease, cancer, dementia, death) ([Barzilai 2016](https://doi.org/10.1016/
- - **Verification before publication.** Every claim needs a verified source. Anything unverified is marked [to verify] and is never used as a basis for a recommendation.
- Every factual claim cites a source whose DOI was checked on Crossref (see `founding-council-r2.check.tsv`), or an official link. Anything not confirmed by a primary source is marked **[to verify]**.
- | **Cochrane** | Thousands of volunteer authors grouped in topic review groups, plus methods groups and a central editorial/methods team that writes the Handbook ([Cochrane Handbook](https://training.cochrane.org/handbook)). Protocols are published before reviews | Mandatory prot
- | **Campbell Collaboration** | Cochrane's sister for social policy, education and crime; coordinating groups by field ([Campbell](https://www.campbellcollaboration.org/)) | Same protocol → review → editorial model | Small volunteer base, slow output **[to verify]** | Proof the Co
- | **CERN / LHC collaborations (e.g. ATLAS)** | Thousands of authors; analysis groups by physics topic; a collaboration board; every paper lists the whole collaboration ([ATLAS authorship policy](https://twiki.cern.ch/twiki/bin/view/AtlasProtected/AuthorShipCommittee) **[to verify
- | **Arc Institute** | Non-profit with core investigators plus shared "technology centres" serving all labs ([Arc Institute](https://arcinstitute.org/)) | Institute-level review **[to verify]** | Young; untested at scale | Shared platform services separate from research groups (ma
- | **Linux kernel** | Hierarchy of subsystem maintainers listed in a MAINTAINERS file; patches flow up to Linus Torvalds ([kernel docs](https://www.kernel.org/doc/html/latest/process/submitting-patches.html)) | Review by subsystem maintainer, signed-off-by chain | Maintainer burno
- The values are from EMPA-REG OUTCOME ([Zinman 2015](https://doi.org/10.1056/NEJMoa1504720)) as cited in the R1 brief; the population numbers are illustrative and must be re-extracted **[to verify]**. Validation rules to add: `ci_low ≤ estimate ≤ ci_high`; ratio measures > 0; `arm
- - Non-cardiovascular death: both classes cut it by about 10 % in RCTs (GLP-1 RA RR 0.90, 0.81–0.99) ([Diabetes Metab Syndr 2024](https://www.sciencedirect.com/science/article/abs/pii/S1871402124000043) **[to verify DOI]**).
- **The gap.** No single, pre-registered analysis pools **all** placebo-controlled outcome trials of **both** classes on a **harmonised set of aging outcomes** (dementia, frailty/function, cancer, fracture, serious infection, non-cardiovascular death) with one outcome definition ea
- Every DOI below was checked on Crossref on 2026-10-19 (`founding-council-r3.check.tsv`, same folder). Official links were checked to load (HTTP 200) the same day. Anything not checked is marked [to verify].
- - **Proposed Astra rule:** a proxy may be converted into years only if trial-level R² ≥ 0.8 (lower 95% bound ≥ 0.6) across ≥ 5 randomised trials of ≥ 2 intervention classes. These cut-offs are an Astra convention for debate, not a published standard [to verify against regulatory
- **TAME.** The TAME trial of metformin was designed to use a composite of age-related diseases as its primary outcome, precisely because no biomarker is validated ([Barzilai 2016](https://doi.org/10.1016/j.cmet.2016.05.011)); its biomarker workgroup proposed blood markers as secon
- **Fundraising preconditions.** Funders of aging research reward credible, open outputs. What makes Astra fundable: (1) the public outcome table with its code; (2) a validated, reproducible engine; (3) at least one result reviewed outside Astra; (4) a governance page (conflicts of
- Two caveats. EXSCEL lists only 18 infection-class terms against ~100–250 in trials of similar size. Its posted SAE table also lists only 2 cardiac terms, in a cardiovascular outcome trial, so the table looks partial (endpoint events probably excluded from the SAE listing), not di
- | Persistent physical disability | Not reported in any of the 27 [to verify by full-text screen] | – | Not estimable from this set |
- | ASPREE (McNeil 2018, 10.1056/NEJMoa1800722; 10.1056/NEJMoa1803955) | ≥ 70 (≥ 65 for US Black and Hispanic), free of CVD, dementia, physical disability | Disability-free survival curve (death, dementia, persistent physical disability), median 4.7 y [to verify] | Matches our endp
- | VITAL (Manson 2019, 10.1056/NEJMoa1809944) | US men ≥ 50, women ≥ 55, free of cancer and CVD | Invasive cancer and major CV events, curves | Right age band inside a wider one; age-specific curves [to verify]; no dementia or disability |
- | WHI estrogen + progestin (2002, 10.1001/jama.288.3.321) | Postmenopausal women 50–79 | CHD, stroke, breast cancer, hip fracture, death; age-stratified rates [to verify] | Women only; 1990s baseline risk |
- **Proposal.** None of the four prints the composite for 60–69. Keep 60–69 composite **"not estimable"**, as decided. A **component-only** 60–69 column is allowed, filled only where a printed result for the 60–69 age band exists (a curve or table restricted to, or stratified at, 6
- | C. Utility decrements (EQ-5D catalogues) | QALY-style decrement per event | Standard in health economics | Different instruments give different values; licensing of some catalogues [to verify] |
- 1. **Full-text access** for supplements and appendices of the 27 trials: most main papers are NEJM/Lancet; several supplements are free, some are not [to verify per trial]. Request: an institutional library proxy or a named academic sponsor (aligned with G10) — no cost estimate y
- 2. **Individual participant data platforms** (Vivli, YODA) for transportability v1: applications require a named academic investigator and a protocol [to verify current terms]. Not needed before wk 12; flag now because approval takes months.
- Every factual claim cites a DOI checked on Crossref (`founding-council-r5.check.tsv`, 15 DOIs, all resolved) or an official web page. Figures marked [to verify] were not re-read in the full text during this round.
- - **Our biggest evidence risk is a wrong number that looks sourced, not a missing paper.** Language models invent references at high rates: 55% of GPT-3.5 citations and 18% of GPT-4 citations were fabricated in one test (Walters 2023), and 47% of 115 references were fabricated in
- | **Fabricated references from language models** | GPT-3.5: 55% of 636 citations fabricated; GPT-4: 18%. Among real citations, 43% (GPT-3.5) and 24% (GPT-4) had substantive errors [to verify exact denominators] | Walters & Wilder 2023, 10.1038/s41598-023-41032-5 |
- | **Data extraction errors (humans)** | Single extraction gave about 22% more errors than double extraction, and was faster [to verify relative figure] | Buscemi 2006, 10.1016/j.jclinepi.2005.11.010 |
- | | Errors found in 20 of 34 Cochrane reviews checked, mostly with little effect on conclusions [to verify] | Jones 2005, 10.1016/j.jclinepi.2004.11.024 |
- | | Meta-analyses using standardised mean differences: in 10 of 27, the pooled result could not be reproduced within 0.1 from the trial reports [to verify] | Gøtzsche 2007, 10.1001/jama.298.4.430 |
- | | Methodological review: extraction error rates are high and vary widely across studies; double extraction and training reduce them [to verify range] | Mathes 2017, 10.1186/s12874-017-0431-4 |
- | | Retracted articles keep being cited after retraction, mostly without mentioning it [to verify proportion] | Bar-Ilan & Halevi 2017, 10.1007/s11192-017-2242-0 |
- | **Spin and overclaiming** | 72 RCTs with a non-significant primary outcome: spin in 58% of abstract conclusions and 18% of titles [to verify title figure] | Boutron 2010, 10.1001/jama.2010.651 |
- | **Sponsor bias** | Industry-sponsored studies report favourable efficacy results and conclusions more often than non-industry studies (risk ratio ≈ 1.3) [to verify exact RR] | Lundh 2017, 10.1002/14651858.MR000033.pub3 |
- | **Redundant or conflicted reviews** | Production of systematic reviews and meta-analyses grew much faster than trials; most are redundant, misleading or conflicted [to verify the quoted share] | Ioannidis 2016, 10.1111/1468-0009.12210 |
- | F3 | Retracted, corrected or expression-of-concern source | Weekly job: query Crossref for update notices (`update-to`, Crossmark) and the Retraction Watch data now distributed by Crossref, for every DOI in `data/` [to verify field names and access terms] | A9 (job), Literature
- - **Journals (COPE).** The COPE retraction guidelines distinguish retraction (findings unreliable, through misconduct or honest error), correction (a small part is wrong), and expression of concern (investigation pending). Retraction notices stay linked to the article; the articl
- - **Crossref and Crossmark.** Publishers register updates (corrections, retractions) as metadata linked to the original DOI, so any reader or machine can see the current status. In 2023 Crossref acquired the Retraction Watch database and opened it (https://www.crossref.org/blog/n
- - **Three lines model (Institute of Internal Auditors, 2020).** Management runs operations and owns risk (first line); risk and compliance functions advise and monitor (second line); internal audit gives independent assurance and reports directly to the governing body, not to man
- - **Cochrane.** Cochrane's editorial process is separate from its funders, and its conflict-of-interest policy bars commercially funded reviews and limits authors with relevant conflicts (Cochrane conflict-of-interest policy, https://www.cochranelibrary.com/editorial-policies [to
- - **Sponsor versus regulator (FDA).** The sponsor runs the trial; the FDA inspects independently (Bioresearch Monitoring programme) and decides on approval (https://www.fda.gov/science-research/clinical-trials-and-human-subject-protection/bioresearch-monitoring-program-bimo [to v
- - **CERN internal review.** Large experiments such as ATLAS and CMS review every paper internally before submission, through review committees drawn from outside the analysis team [to verify, no DOI found this round]. The point for us: the reviewers are colleagues, but not from t
- Official pages (not DOI, [to verify] current URLs): COPE retraction guidelines, https://publicationethics.org/guidance/guideline/retraction-guidelines; Crossref and Retraction Watch, https://www.crossref.org/blog/news-crossref-and-retraction-watch/; Crossmark, https://www.crossre
- **Sources.** Every DOI was checked on Crossref on 2026-10-19. The results are in `founding-council-r6.check.tsv` (16 resolved and 1 failed, which is not cited). Trial counts come from the R4 brief (ClinicalTrials.gov API v2). Claims I could not check are marked [to verify].
- *Attack.* Outcome reporting bias is common and changes conclusions. [Kirkham 2010](https://doi.org/10.1136/bmj.c365) found it suspected in at least one trial in about a third of Cochrane reviews. In a share of meta-analyses, it was enough to change a significant result [exact pro
- - [Riveros 2013](https://doi.org/10.1371/journal.pmed.1001566) found SAEs more completely reported on ClinicalTrials.gov than in journals [proportions to verify];
- 1. **Medical-advice and legal exposure.** A public "best anti-aging intervention" ranking will be read as advice. Proposal: a fixed disclaimer, no dosing and no individual recommendations; entries for prescription drugs say "discuss with a doctor; evidence in sick populations". W
- 4. **Licences of extracted data.** CT.gov records are public. Journal tables and digitised curves may be under publisher copyright. Extracted numbers are generally treated as facts, but bulk republication of supplements is not [to verify]. Proposal: store only numbers plus a sour
- 7. **Sources**: every DOI checked on Crossref (`dossier.check.tsv`). Anything not yet checked, or any number not yet read in the primary paper, is marked `[to verify]`.
- No number appears here without a source. Where secondary sources disagree, both values are given and the item stays `[to verify]` until the primary paper is read.
- | Thompson 2018 | Multi-tissue | 582 (elastic net, all sites); 274 (conserved sites) | `[to verify]` | C57BL/6 multi-tissue data, as described by Simpson 2023 | Leave-one-batch-out: r = 0.82, median error 2.5 months; conserved-site clock r = 0.68, 3.8 months | Thompson 2018 Table
- Age data are almost all from C57BL/6 (B6), mostly males. No UM-HET3 reference curve found `[to verify]`.
- - No published same-mouse test-retest coefficient (ICC or CV) found. `[to verify]`
- Take-home: in B6 mice, GTT AUC is a poor ageing readout on its own. It tracks fat mass, rises then falls with age, and hides real age changes (lower insulin sensitivity and clearance) seen in clamp studies (Ehrhardt 2019; Marmentini 2021). Use it only with body composition measur
- No standard reference. Age data are mostly from male C57BL/6 (B6). No UM-HET3 reference curve found `[to verify]`.
- No standard reference. Age data are mostly from male C57BL/6 (B6). No UM-HET3 reference curve found `[to verify]`.
- - Scores were reliable across repeated test sessions (de Fiebre 2006); no reliability coefficient was in the parts read `[to verify]`.
- - **Repeat testing.** Mice remember the maze, so retesting the same mouse is not a fresh measure. Longitudinal use needs a new escape position or new cues `[to verify]` (no source read today).
- - Normal-hearing adults, standard frequencies and EHF up to 16 kHz, two earphone types (Schmuziger 2004). n and ages `[to verify]`.
- - In plain English: one threshold moving by **10 dB or less** between two tests is within noise. A shift of **≥10 dB at one frequency, or ≥5 dB averaged over several adjacent frequencies, repeated on a second test**, is the usual signal of real change `[to verify]` exact criterio
- Hearing is itself the function. Age-related loss starts at the highest frequencies and spreads down. `[to verify]` population norms (e.g. ISO 7029) before quoting numbers.
- - Every DOI checked on Crossref; anything unchecked is marked `[to verify]`.
- - **intervention:** acarbose; target intestinal α-glucosidase (maltase-glucoamylase MGAM / sucrase-isomaltase SI); ChEMBL target id `[to verify]`
- - **intervention:** dasatinib + quercetin (D+Q) regimen; dasatinib target SRC/BCR-ABL family, quercetin multi-target; recorded as the regimen, ChEMBL per-compound ids `[to verify]`
- - **grade:** direction supported at E2-equivalent (one high-SYRCLE causal mammalian test); **size ungraded** (`no_mapping`). Two further clean-abstract sources not opened for text: doi:10.1111/acel.70759 (heart), doi:10.14336/AD.2025.0976 (arteries) — `[to verify]`, available to
- - **Open `[to verify]`:** acarbose ChEMBL target id (M-003); D+Q per-compound ChEMBL ids (M-005); the two extra senolytic sources (M-005).
- - **version:** v2. **Diff from v1:** ChEMBL ids replace `[to verify]`; SYRCLE count and domain list replace "< 5" (F2); Q1 added (F1). Size **0.442 [0.402, 0.485]**, direction and status unchanged.
- - **excluded / not used:** Zhu 2015, doi:10.1111/acel.12344: **not used**, superseded draft prior, abstract read (healthspan only). Not an `X-LS` exclusion. Two clean-abstract sources not opened (doi:10.1111/acel.70759, doi:10.14336/AD.2025.0976) stay `[to verify]`.
- - **version:** v2. **Diff from v1:** per-panel marker ratios added (from #228, check #230); NA4.1 lines and dilution added; no single multiplier; grade E2-equivalent → **E1** with the domain list (F2); ChEMBL ids replace `[to verify]`; Zhu 2015 re-recorded as "not used"; limits 1
- Open `[to verify]`: the two unopened senolytic sources (M-005).