A GLP-1 gets a lifespan curve — and the clearance of senescent cells turns out to be an immune-checkpoint problem
The headline result is that semaglutide, started at 20 months of age, extended median lifespan in female mice from 742 to 834 days while improving muscle and cognitive function — the first lifespan curve for a GLP-1 receptor agonist, and, in a matched head-to-head, not fully reducible to reduced calorie intake. It is a mouse study in one sex at one dose, and it says nothing yet about human lifespan. Alongside it, senescence research converged on immune evasion: PD-L2 is a senescence-associated checkpoint that lets senescent cells persist, and blocking it restores insulin sensitivity in aged mice, while senescent microglia were shown to damage oligodendrocytes through a single secreted factor, DLK1. Two genome-maintenance papers make damage look like a modifiable axis in both directions — caloric restriction lowered genome-wide somatic mutation burden in mice, and platinum chemotherapy in children left livers carrying an adult-sized mutation load. Measurement work continued to complicate population-level aging metrics, and two widely shared 'longevity' claims this month do not survive reading the primary source.
Median lifespan of female mice started on semaglutide at 20 months of age versus untreated controls (~12% extension) — the first lifespan curve for a GLP-1 receptor agonist. Mouse data only; no human lifespan claim is supported.
Where the human evidence is
Count by statusEvidence-stage map
Evidence stage × investigator promise (1–10)Senescence: immune evasion, not just production
PD-L2 is a senescence-associated immune checkpoint; PD-L2 knockout or anti-PD-L2 antibody reduces senescent-cell burden and restores insulin sensitivity in aged mice
PD-L2 is upregulated in isolated senescent human cells and with aging, and senolytics preferentially remove high-PD-L2 senescent cells. Old PD-L2 knockout mice accumulate fewer senescent cells and have greater insulin sensitivity and grip strength than old wild-type mice, and anti-PD-L2 therapy restored insulin sensitivity in aged wild-type animals. Mechanistically this reframes senescent-cell persistence as immune evasion via a checkpoint rather than simple overproduction — the same logic as the macrophage-efferocytosis work covered on Sept 6, but with a druggable ligand and an existing clinical-grade modality. Two cautions before this reaches practice: checkpoint blockade in older adults carries real autoimmune toxicity risk and has never been trialled for an aging indication, and the human data here are correlative (soluble PD-L2 tracking with age and falling after senolytic exposure), not interventional.
Read the paperTelomere-shortened senescent microglia secrete soluble DLK1, causing hypomyelination and neuronal dysfunction in the aging brain
DLK1 was identified as a senescence-associated ligand in senescent human iPSC-derived microglia and then validated in vivo: telomere-shortened mice showed lipofuscinosis, hypomyelination, microglial atrophy and cognitive deficits, with single-nucleus RNA-seq showing accelerated glial aging. Soluble DLK1 rose in CSF of telomere-shortened and naturally aged mice and was abolished by microglial depletion, establishing microglia as the source; sDLK1 impaired oligodendrocyte function and altered neuronal activity. Age-related DLK1 increases were also found in human brain single-cell datasets. This nominates a single secreted mediator — and therefore a potential CSF biomarker and neutralisation target — for the senescent-microglia arm of brain aging, which matters because the one completed human senolytic brain trial (D+Q in mild Alzheimer's) was biomarker-null. Still entirely preclinical; no anti-DLK1 agent exists for this indication.
Read the paperGenome maintenance is modifiable — in both directions
Caloric restriction reduces genome-wide somatic mutation burden across tissues in mice, including the ubiquitous SBS5 process
Using high-fidelity duplex sequencing of bulk liver, bulk kidney, hepatocytes and cerebellar neurons, CR lowered both substitution and indel burdens in mice, with effect size varying by tissue and cell type, and specifically decreased activity of SBS5 — the clock-like mutational process responsible for most mutations in mammals. Counterintuitively, the reduction was largest in transcriptionally inactive regions, which argues against a simple transcription-coupled-repair explanation. The significance for geroscience is that somatic mutation accumulation, long treated as an irreversible substrate of aging, behaves as a dose-responsive and diet-modifiable axis. It does not follow that CR-mimetics lower mutation burden, that the effect is causal for lifespan, or that any of this is measurable in humans today.
Read the paperPlatinum chemotherapy in childhood leaves livers with an adult-sized mutation burden (~2,200 mutations per liver sample), dose-proportional to exposure
Children treated for hepatoblastoma with cisplatin and/or carboplatin had normal liver tissue sequenced by NanoSeq and compared with untreated children, differently-treated children and fetal liver. Platinum exposure produced on average ~2,200 mutations per liver sample — a burden normally seen in adult livers — with load increasing proportionally to exposure (cisplatin alone < cisplatin plus carboplatin). Mutations hit cancer genes and genes tied to long-term liver metabolism, and hepatocytes carried far more mutations than blood cells despite systemic dosing, implying tissue-specific handling or repair. The authors and the accompanying Perspective argue for survivorship monitoring beyond the third decade. Important limits: this is mutation burden, not observed cancer or organ failure, the cohort is small and single-indication, and the authors explicitly decline to claim these cells will become malignant.
Read the paperMeasurement: organ-level structure and the limits of population averages
PathStAR quantifies structural aging from routine histopathology across 25,306 biopsies, 40 tissues, 970 donors — and finds nonlinear, organ-specific trajectories
PathStAR scores tissue structural aging from standard histology without being trained to predict chronological age, which avoids the circularity that afflicts most age predictors. Applied to GTEx-scale material, it shows organ structural aging is nonlinear and tissue-specific: vascular tissue accelerates early, uterus and vagina late (around menopause), digestive and male reproductive organs biphasically. Accelerations share a signature of increased inflammation with reduced energy production, repair and quality control. As proof of concept it recovers the known nonlinear ovarian decline — fertility loss in the 30s, menopause in the 50s — that matched bulk transcriptomic and methylation profiles from the same samples fail to detect, which is a pointed result for anyone using a single blood methylation clock as an organ-level readout. Cross-sectional, post-mortem, donors aged 21–70 only; not a clinical test.
Read the paperEight-year longitudinal multi-omics in 335 women: 5,061 genes and 181 metabolites change with age, but individual trajectories routinely diverge from the population trend
Repeated whole-blood transcriptomes and metabolite panels over eight years show that molecular aging is dynamic, context-dependent and highly individual: longitudinally variable genes were cell-type specific and enriched for cardiometabolic and neurodegenerative pathways, and trajectories were shaped by genetics, circadian timing, seasonality and pollutant exposure. The practical consequence is methodological — cross-sectional population averages, including the ones underlying most aging clocks, can misrepresent what is happening in a given individual, and a single timepoint cannot distinguish a fast ager from someone sampled at an unrepresentative moment. Female-only cohort, whole blood only, 335 participants; this constrains interpretation of biomarker panels rather than offering a replacement for them.
Read the paperClaims that do not survive the primary source
The viral 'caffeine flips an ancient longevity switch' coverage traces to a 2025 fission-yeast paper, not new human or mammalian data
A wave of September 2026 coverage described caffeine activating AMPK as an aging intervention. The underlying experiment is in Schizosaccharomyces pombe, where caffeine accelerated mitotic division and benefited chronological lifespan through AMPK pathway components (Ssp1, Ssp2, Amk2); the authors' own claim is that pharmacological AMPK targeting merits study beyond yeast. There is no mammalian lifespan data, no human data, no dose translation, and the paper is from mid-2025 rather than this month. AMPK remains a legitimate geroscience target; that is not the same as evidence that habitual caffeine intake slows human aging, and observational coffee–mortality associations cannot be recruited as mechanism. Flagged here because patients will arrive with this story.
Read the paperDelayed-release calcium-AKG associated with 1.8-year lower epigenetic 'Age Residual' — in a self-selected, supplement-buying cohort using a proprietary clock
Across 4,260 'health enthusiasts' who purchased saliva methylation tests between 2020 and 2025 and completed lifestyle questionnaires, 84 supplements were tested against a proprietary 9-CpG clock (mean absolute error 5.4 years). Delayed-release calcium-alpha-ketoglutarate plus vitamins showed the strongest association — an average 1.8-year lower Age Residual, surviving adjustment for age, sex, smoking and health status. What this is not: randomised, blinded, or independent of the test vendor and supplement market; 71% of the cohort used supplements, self-selection and healthy-user bias are unquantified, and the outcome is a proprietary clock rather than function or events. Note the trajectory of the claim — the 2021 Rejuvant report described an ~8-year reduction; this larger analysis puts the associated difference at 1.8 years. The randomised AKG trials (e.g. ABLE) remain the evidence that would matter.
Read the paper