Single-cell patterns of monocyte migration is a predictive biomarker of aging and frailty

Introduction: Aging is defined as a process of functional decline across multiple organs and tissues, and is a primary factor driving numerous chronic diseases. However, there is a high heterogeneity among all individuals that leads to inconsistencies in aging phenotypes and risks. Biological age, which is defined as the long-term changes that determine the functional health trajectories of individuals, is a better predictor of risks than chronological age and various clinical metrics have been measured based on frailty and resilience status. Frail individuals suffer from increased vulnerability due to physiological decline and disability to maintain functional integrity after exposure to stressors and perturbations. Recent studies have identified several drivers of age-associated dysfunctions at the molecular and clinical levels, but sensitive meso-scale cell level biomarkers remain understudied. Biophysical properties at the single-cell level were discovered to have a strong correlation with age. Here, we combine single-cell behavior patterns with machine learning and deep neural network models to predict age and frailty status with high accuracy.

Materials and

Methods: Human peripheral blood mononuclear cells (PBMCs) were isolated from young, old and frail donors and CD14+ monocytes were further isolated. We embedded primary monocytes into collagen gels and performed time-lapsed confocal microscopy with intervals of 2 minutes for 3 hours. We further profiled the responses of cells to pro-inflammatory signals using IL-6, DNA fragments, and LPS. Spatiotemporal behaviors were collected and profiled using an automatic pipeline, which used deep learning-based segmentation using CellPose 10, feature extraction using our customized pipeline, single-cell tracking based on the Crocker-Grier algorithm using the Trackpy Python package. ~200 dynamic behaviors and colocalization features have been measured using our customized pipeline. Unsupervised K-means clustering has been performed to define 12 distinct behavior patterns. For age group classification, we developed scTRAIT, which integrates movement trajectories, colocalization effects, morphodynamics and perturbation effects to learn spatiotemporal behavioral patterns to accurately classify young, old and frail individuals.

Results, Conclusions, and Discussions: Behaviors of the monocytes from young, old and frail donors were mapped onto a reduced dimensional space using UMAP, and we identified a unique distribution of each age groups (Fig A). We defined 12 distinct patterns of behavioral clusters that share similar movement patterns (Fig B, C). We found that these patterns are distinct not only across age groups but also the responses to inflammatory stimulation by IL-6, DNA fragments, and LPS are unique (Fig D). There was a clear linear correlation between chronological age and movement speed for both old and young groups, and the correlation coefficients varied based on the type of perturbation. Finally, we built scTRAIT, which utilizes all single-cell behavior information and behavioral shift due to perturbations to predict human-level age groups and frailty status. We obtained the best results for classification of young and old with 89% accuracy (Fig F) and frailty status with 91% accuracy (Fig G) using our scTRAIT model, which has the highest accuracy compared to other models.

We identified significant differences in monocyte behaviors and responses to pro-inflammatory signals between age and frail groups. These spatiotemporal behaviors were fed into our deep neural network-based model, scTRAIT and successfully predicted human-level ages and frailty status. This study suggests that single-cell behavior dynamics encode age and frailty information and serve as new cell-level biomarkers.

Acknowledgements (Optional):