Clocks split by sex — and the NAD+ biomarker fails its own audit
Measurement was the week's real subject. Nature Medicine published 38 sex-specific biological aging clocks across 15 organ systems and showed that organ-level aging liability, its genetic architecture and its links to cardiometabolic and mental traits are configured differently in women and men — meaning a sex-pooled clock is measuring against the wrong normative reference for roughly half of any clinic's panel. Cell released LongevityBench, the first open benchmark for whether AI systems can actually interpret aging biodata, and found omics-based age prediction is the hardest task for every frontier model tested regardless of scale. Against that, the NAD+ story tightened in an uncomfortable direction: a new Japanese cohort links lower whole-blood NAD+ to frailty, while a rigorously validated LC-MS analysis across seven independent human cohorts finds whole-blood NAD+ essentially flat with age and unmoved by lifestyle — moving only with nicotinamide riboside. On the mechanism side, two Nature Aging papers put APOE4 and the aged cortex under new resolution, and a 3,119-person Chinese cohort argues cognitive resilience is a second, independent and equally weighted axis of dementia risk alongside pathology.
Per-SD hazard ratios for incident Alzheimer's dementia from cognitive resilience (0.51, 95% CI 0.48–0.55) versus pathological burden (2.50, 95% CI 2.30–2.72) in 3,119 adults followed a median 13.7 years — two independent axes of comparable weight that also interact multiplicatively. Observational; resilience is a statistical residual, not a measured biological construct.
Where the human evidence is
Count by statusEvidence-stage map
Evidence stage × investigator promise (1–10)Measurement: clocks split by sex, and AI gets a benchmark
Thirty-eight sex-specific biological aging clocks across 15 organ systems show female and male aging liability is configured differently at the genetic, proteomic and clinical level
The authors built 38 sex-stratified clocks across 15 organ systems and first demonstrate why sex-stratified training matters: it is what makes a sex-appropriate healthy normative reference possible at all. Key genetic parameters and Mendelian randomisation indicate that organ-specific aging liability — and its relationships to cardiometabolic, endocrine and mental traits — differs between females and males, and proteomic analysis identifies distinct organ-resolved synaptic, immune, vascular and metabolic networks tracking each. In longitudinal survival analysis the clocks predict systemic disease and all-cause mortality in a sex- and organ-dependent manner, and the brain clock showed sex-dependent associations with cognitive decline trajectory inside a preclinical Alzheimer's trial. This is a different and larger result than the sex-stratified clinical clocks covered on Aug 30 (Nature Aging, 172 measures): that paper was phenome-based and described divergent midlife trajectories, this one is organ- and omics-resolved and carries survival and trial-embedded validation. The authors are explicit that sex-pooled and sex-interaction models remain valuable; the claim is not that pooled clocks are invalid but that they obscure sex-dependent signal. Not a clinical test, and the normative references are cohort-specific.
Read the paperLongevityBench: 17 tasks across five aging biodata domains, 18 frontier AI systems — and omics-based age prediction is the hardest task for every model regardless of scale
Two decades of human aging data across methylation, transcriptomic, proteomic and clinical modalities have never been assembled into a test of whether AI systems can actually interpret them. LongevityBench is an open 17-task suite spanning five biodata domains, used here to assess 18 frontier AI systems from six developer teams. No single model dominates, and omics-based age prediction remains the hardest task irrespective of model scale — a finding worth sitting with, given how much clinical longevity practice now routes through algorithmic age estimates. The authors then fine-tuned compact multitask Longevity-LLMs (0.6B–9B parameters) that matched or exceeded far larger frontier systems, and released the benchmark, models and an agentic research interface. Read it as infrastructure, not a clinical claim: the work comes from Insilico Medicine, a commercial longevity-AI company benchmarking a field it competes in, and a benchmark's task selection is itself a scientific choice. Nature ran a companion news feature on AI aging clocks the same week.
Read the paperNAD+: the blood biomarker does not behave the way the field assumed
Whole-blood NAD+ is lower in frail community-dwelling older adults (n=529, 6.0% frail) — the first sizeable human frailty association
In 529 community-dwelling Japanese adults aged 65 and over (54.3% women) from the 2025 ILSA wave, 32 participants (6.0%) met revised Japanese Cardiovascular Health Study frailty criteria, and median whole-blood NAD+ concentration was lower in that group. Multivariable linear and logistic models adjusted for age, sex and fasting time, with restricted cubic spline, Firth penalised-likelihood and haematological sensitivity analyses supporting the association. This is useful because human evidence linking circulating NAD+ to a functional phenotype has been thin — but it is cross-sectional with only 32 frail participants, so the confidence intervals are wide, reverse causation is unaddressed, and no supplementation inference follows. Pair it with the item below before drawing any conclusion about measuring NAD+ in practice.
Read the paperWhole-blood NAD+ does not decline with age or shift with lifestyle interventions across seven independent human cohorts — only nicotinamide riboside moves it
The premise underneath most consumer NAD+ testing and much NAD+ precursor marketing is that blood NAD+ falls with age and rises with healthy living. Using a rigorously validated UHPLC–high-resolution MS method built to account for real-world analytical variability, the authors quantified NAD+ across seven independent human cohorts and found whole-blood levels remarkably stable with age and across lifestyle interventions — while responding, as expected, to nicotinamide riboside supplementation. Their stated conclusion is that this challenges the utility of blood NAD+ as a biomarker of ageing or of lifestyle. The methodological point is the substance: much of the reported age-related decline in the prior literature is plausibly assay variability rather than biology. This does not refute tissue-level NAD+ decline, does not refute NAD+ precursor pharmacology, and does not negate the frailty association above — but it does mean a single blood NAD+ value should not be sold to a patient as a biological-age readout, and that a post-supplement rise in blood NAD+ demonstrates absorption, not benefit. Flagged as challenged rather than new because the paper is from May; it is included because this week's frailty paper cannot be read without it.
Read the paperNeuro-longevity: APOE4 at the barrier, and a hidden proteome in the aged cortex
Astrocyte-derived fibronectin mediates APOE4-driven blood–brain barrier dysfunction, nominating FN1 as a tractable vascular target
The authors identify astrocyte-derived fibronectin as a key driver of blood–brain barrier dysfunction in Alzheimer's disease, mechanistically linking APOE ε4 and amyloid pathology to vascular damage and highlighting FN1 as a potential therapeutic target. The significance for a longevity practice is that it gives the APOE4 vascular phenotype a specific extracellular-matrix mediator rather than a generic 'BBB leakiness' description, which is what a druggable or monitorable target requires. Preclinical and mechanistic; there is no FN1-directed agent in human trials for this indication, and nothing here changes APOE4 risk counselling or current management. Note this sits alongside a separate Nat Commun paper this cycle on APOE genotype remodelling the astrocytic lipid-droplet proteome — the astrocyte is becoming the convergence point for APOE4 biology.
Read the paperA microprotein atlas of the human frontal cortex identifies 1,067 unannotated microproteins, one MKKS-derived microprotein required for normal microglial mitochondrial respiration
Microproteins of roughly 100–150 amino acids are largely absent from reference proteomes and have been effectively invisible to brain and aging research. This atlas catalogues 1,067 unannotated microproteins in human frontal cortex with strong spectral grades, a subset of which are differentially expressed in Alzheimer's disease independently of their canonical parent genes — including an MKKS-derived microprotein required for normal microglial mitochondrial respiration. The implication is that a measurable slice of the aged brain's regulatory biology has been missing from every proteomic aging study run to date, including the proteomic clocks now entering clinical use. Descriptive human post-mortem work with one functional validation; no biomarker or intervention follows yet, and microprotein quantification is not clinically available.
Read the paperSenescence: a proteostasis signature, and a roadmap the field agreed on
Quantitative proteomics of replicative senescence: widespread chromatin protein depletion, loss of cytoplasmic translation machinery, mitochondrial protein insolubility and compromised autophagy–proteasome flux
Mass-spectrometry proteomics tracked dynamic proteome changes at multiple levels through the progression of replicative senescence and found the changes are coordinated rather than piecemeal: widespread protein depletion on chromatin, depletion of the cytoplasmic translation machinery, increased insolubility of mitochondrial proteins, compromised autophagic and proteasome activity, remodelled ubiquitin linkages and depletion of ubiquitin E3 ligases. Comparison against other pathophysiological cellular states yields a distinctive senescent signature shaped by the proteostasis network. This matters directly to the field's definitional problem flagged on Aug 30 — where only 173 of 1,250 genes across four standard senescence gene sets were corroborated by two or more resources — because a proteostasis-level signature is a candidate discriminator that transcript panels have not provided. In vitro replicative senescence in cultured cells only; it is a resource, not evidence that any senolytic acts through this axis.
Read the paperSENESCENCE2030 consensus roadmap: functional classification of senescent states, standardised biomarkers, and regulatory frameworks named as the blocking constraints on precision senescence medicine
Researchers, clinicians, industry and policy stakeholders at the SENESCENCE2030 Annual Conference and Industry–Academia Workshop (Coimbra, May 2026) set out what has to be solved before senescence biology reaches practice: functional classifications of senescent states, standardised and clinically actionable biomarkers, precision senescence medicine, translational and regulatory frameworks, and international collaboration. For anyone fielding patient questions about senolytics, the useful signal is what the field's own consensus does not claim — there is no agreed biomarker to select patients, no agreed definition of the target cell population, and no regulatory pathway for an aging indication. A consensus roadmap carries no experimental weight of its own and reflects the priorities of the participants; treat it as an honest statement of where the bottlenecks are, not as evidence about any compound.
Read the paperContext and pipeline: the economics of geroscience, and a klotho claim to watch
Reframing diseases by when they occur across the lifespan: aging-related disease now dominates global burden in every income group — and uniquely shows increasing returns
Categorising global disease burden by life stage rather than by organ system shows aging-related diseases dominating in every income group, including low-income settings usually modelled around infectious and maternal burden. The analytically interesting claim is the second one: aging-related diseases uniquely exhibit increasing returns, meaning the more of the burden that is addressed, the larger the marginal gain — because these conditions cluster in the same individuals and span many years, so removing one does not simply reveal the next. The accompanying News & Views frames the policy consequence as a pivot from treating disease to preserving health across all life stages. This is an economic modelling analysis built on burden estimates with their own uncertainty, and 'increasing returns' is a property of the model's structure as much as an empirical discovery — but it is the clearest published argument for why geroscience-directed spend should behave differently from disease-by-disease spend.
Read the paperAvaí Bio / Austrianova present encapsulated-cell α-klotho production data at the Second Annual Klotho Conference (Miami, Sept 18–19 2026) — conference presentation, no peer-reviewed data, microcap promotion context
The approach is a cell-encapsulation platform intended to sustainably restore circulating α-klotho rather than deliver recombinant protein or gene therapy, presented via the Klothonova joint venture at the Healthy Longevity Clinic–hosted conference this week. It is listed here because klotho is among the most asked-about interventions in longevity practice and because patients will encounter the coverage, not because the data support anything. What is on record is a conference presentation and press releases from a microcap issuer, with no peer-reviewed publication, no registered efficacy trial and no human outcome data; several of the circulating items are stock-promotion pieces rather than science reporting. The underlying klotho evidence base remains what it was: strong mouse and non-human primate cognition data (Nature Aging 2023), robust epidemiological associations, and no controlled human trial demonstrating that raising circulating α-klotho improves a clinical outcome. Unverified beyond the fact that the presentation occurred.
Read the paper