Multidimensional phenotyping predicts lifespan and quantifies health in Caenorhabditis elegans

Ageing affects a wide range of phenotypes at all scales, but an objective measure of ageing remains challenging, even in simple model organisms. To measure the ageing process, we characterized the sequence of alterations of multiple phenotypes at organismal scale. Hundreds of morphological, postural, and behavioral features were extracted from high-resolution videos. Out of the 1019 features extracted, 896 are ageing biomarkers, defined as those that show a significant correlation with relative age (age divided by lifespan). We used support vector regression to predict age, remaining life and lifespan of individual C. elegans. The quality of these predictions (age R2 = 0.79; remaining life R2 = 0.77; lifespan R2 = 0.72) increased with the number of features added to the model, supporting the use of multiple features to quantify ageing. We quantified the rate of ageing as how quickly animals moved through a phenotypic space; we quantified health decline as the slope of the declining predicted remaining life. In both ageing dimensions, we found that short lived-animals aged faster than long-lived animals. In our conditions, for isogenic wild-type worms, the health decline of the individuals was scaled to their lifespan without significant deviation from the average for short- or long-lived animals.Author summary: High dimensional biomedical data are used to quantify health and diagnose diseases. Combining the most informative features collected in the best conditions is crucial for predictive power. Using high-resolution videos and extraction of hundreds of morphological, postural and behavioral features, we characterized the phenotypic evolution of worms as they age. Out of the 1019 features extracted, 896 correlate with relative age. We used machine-learning to predict age and lifespan of individual C. elegans (age R2 = 0.79; remaining life R2 = 0.77; lifespan R2 = 0.72). The quality of these predictions increased with the number of features added sequentially to the model, supporting the use of multiple features to quantify ageing. We evaluated the relationship between ageing and the phenotypic progression. In our conditions, for isogenic wild-type worms, the rate of phenotypic alterations scales with the lifespan of the individuals.