Biological Age Testing

Multiple validated epigenetic clocks. GrimAge and DunedinPACE predict mortality and healthspan. Increasingly used to track intervention efficacy. UPDATE (August 2026): Two peer-reviewed additions this cycle. (1) Nature Aging 2026 (Robinson et al.) evaluated plasma SomaScan proteomic age clocks, including organ-specific clocks, in 17,473 EPIC participants over up to 28 years: a composite global proteomic age gap showed the strongest all-cause mortality association of any clock tested and tracked cardiovascular disease, dementia and several cancers, with organ-specific gaps preferentially associated with cancers of the matching organ. The authors' own caveat is important - mortality prediction was COMPARABLE to, not better than, classical lifestyle risk factors, so incremental clinical utility over standard risk tools is still unproven. (2) A Scientific Reports 2026 proof-of-concept (n=123, monocyte-enriched) found gut microbiome composition predicted DunedinPACE (species-level R2=0.152) but showed NO predictive signal for Horvath, Levine or GrimAge2 - a concrete reminder that "epigenetic age" is not one measurement and that clocks of different generations behave differently. Separately, a Nature Aging 2026 meta-analysis across 17 human tissues (>15,000 profiles) found systemic methylation shifts plus rising methylation variability and molecular disorder with age, identifying PCDHGA1 as a conserved cross-tissue hub. UPDATE (August 9, 2026): Nature Medicine 2026;32(6):2060-2072 (Ding et al., Wyss-Coray lab) extends proteomic aging from organ-level to CELL-TYPE-level resolution: models built on >7,000 plasma proteins in 60,542 individuals estimated the biological age of more than 40 cell types across neuronal, glial, immune, endocrine, epithelial and musculoskeletal lineages. 20-25% of individuals showed accelerated aging confined to a single cell type and 1-3% across ten or more, and the signatures predicted incident disease and mortality over 15 years. Extreme-decile effects are clinically legible: extreme astrocyte aging tripled incident Alzheimer's risk in APOE4 homozygotes (youthful astrocytes lowered it), extremely aged skeletal myocytes carried 12.7-fold higher ALS risk, and in smokers extreme respiratory epithelial aging added 58% to lung-cancer risk beyond smoking alone. A polycellular aging risk score stratified mortality across cohorts and platforms. This is observational cohort work — a risk-stratification framework, not a validated intervention endpoint — and tail hazard ratios will attenuate substantially in unselected practice. For expectation-setting alongside any biological-age result, an npj Aging 2026 modeling study (Efimov et al.) estimated that even if every reversible hallmark were eliminated, somatic mutation alone would cap median human lifespan at 146-194 years, with post-mitotic tissue as the bottleneck; it is a model rather than an observation, but it usefully bounds what any biological-age optimisation programme can be expected to deliver. UPDATE (August 16, 2026): The field has turned its instruments on the clocks themselves, and the result cuts against reporting a single "biological age" number. (1) PEER-REVIEWED: npj Aging 2026 (Arpawong, Crimmins et al.) paired DNA methylation with contemporaneous RNA-seq in 3,227 US Health and Retirement Study participants and ran differential expression against age acceleration for Horvath, Hannum, PhenoAge, GrimAge and DunedinPACE. The enriched biological pathways behind each clock were LARGELY NON-OVERLAPPING - spanning immune activity, metabolism, cell growth and signalling - meaning the five canonical clocks are not noisy estimates of one latent construct and averaging or substituting between them is not defensible. Transcriptomic aging gene scores (TAGS) derived from those clock-specific signatures complemented, and in several cases outperformed, the DNAm clocks for association with age-related morbidity and mortality. Note the paper is posted as an unedited accepted manuscript. (2) PREPRINT - NOT PEER-REVIEWED: a bioRxiv preprint posted 5 August 2026 (Wang, Ionita-Laza et al.) built calibrated prediction intervals and individualised probabilities of accelerated or decelerated aging around three composite and eleven organ-specific proteomic clocks in the UK Biobank Pharma Proteomics Project. For low-accuracy clocks, including many organ-specific ones, apparently extreme age gaps carried little evidential weight at the individual level. Prediction-interval width was itself independently associated with disease risk and mortality (strongest for composite, brain and immune clocks), replicated in Biobank Japan and a Stanford clinical cohort. This is a preprint and must not be cited as established, but its direction is consistent with the peer-reviewed heterogeneity findings above and it is directly relevant to how organ-clock reports are worded for patients. UPDATE (August 23, 2026): The most consequential development yet for how these tests should be interpreted, and it is peer-reviewed. (1) Nature Medicine 2026 (Sehgal, Borrus & Higgins-Chen, Analysis, open access) curated TranslAGE — a harmonized database of 51 public and private longitudinal INTERVENTIONAL studies — and computed a consistent set of 16 prominent epigenetic clocks plus 94 additional DNAm biomarkers across every study. This is the first systematic test of whether DNAm clocks behave the way a surrogate endpoint should: responding when there is genuine clinical benefit and staying stable when there is not. The findings are conditional rather than reassuring. Clocks trained on MORTALITY or PACE OF AGING (generation 2+, reliability-optimized) showed the strongest responses across interventions and the most agreement with one another; first-generation chronological-age predictors were markedly less responsive. Pharmacological and lifestyle interventions drove the largest biomarker movement. Critically, the characteristics of the study population and the study duration were decisive determinants of whether ANY response was detectable — meaning a null clock result can reflect trial design rather than an ineffective intervention. Multi-subscore "explainable" clocks provided mechanistic specificity that single-score clocks could not. In the accompanying News & Views, Steve Horvath frames this as a report card on surrogacy; note his disclosed GrimAge/PhenoAge patent interests and Altos Labs equity when weighing the framing. (2) Nature Medicine 2026 (histological aging signatures) adds an entirely orthogonal modality: deep-learning "tissue clocks" trained on 25,712 whole-slide histopathology images spanning 40 tissue types from 983 GTEx donors, with roughly 4.9 years mean prediction error. Tissue age gaps tracked telomere attrition, subclinical pathology and comorbidity; organ-specific acceleration mapped onto demographic and lifestyle factors; and by integrating paired histology and transcriptomics the authors predicted tissue-specific age gaps FROM BLOOD, validated against eight prevalent diseases including Alzheimer's, stroke and Crohn's in independent cohorts. Two caveats matter clinically: GTEx is a post-mortem donor cohort, and the blood-inferred tissue ages are a derived inference layer rather than a direct measurement of anyone's tissue.

Evidence Scale