Vaudano, A. P.; Pierron, M.; Stojkovic, L.; Membrez, M.; Bourgeois, M.; Neal, C.; Chimen, M.; Verbakel, L.; Cornaglia, M.; Solari, F.; Mouchiroud, L.

Interventions that extend lifespan do not necessarily preserve healthspan, the portion of life spent in good health. This disconnect has intensified interest in biological aging clocks as quantitative proxies of organismal health. However, most existing clocks rely on invasive or endpoint measurements, providing static estimates that capture biological age at a single time point and offer limited insight into aging trajectories - the dynamic patterns through which physiological resilience and functional capacity change within individuals over time. Here we combine standardized, high-frequency imaging of individual Caenorhabditis elegans across the lifespan with machine learning to develop MOSAIC (Modular Organismal Signature of Aging In C. elegans), a non-invasive phenotypic clock that estimates biological age longitudinally at single-organism resolution. Leveraging ~3'750 animals, ~230'000 observations and 29 phenotypic features, MOSAIC predicts biological age with high accuracy and resolves organism-wide aging trajectories at high temporal resolution. Beyond age prediction, MOSAIC decomposes biological age into contributions from distinct physiological modules, enabling mechanistic interpretation of organismal decline. Applying MOSAIC to natural lifespan variation, dietary restriction, longevity mutants and pharmacological interventions reveals that lifespan extension can emerge through distinct, time-dependent phenotypic trajectories rather than a uniform slowing of aging. Interventions with similar effects on longevity produce divergent biological-age trajectories and distinct combinations of younger and older traits, highlighting context-dependent physiological trade-offs. MOSAIC provides a scalable, non-invasive framework to repeatedly quantify biological age across the lifespan and to compare interventions based on how they reshape aging trajectories.