Abstract
"Biological aging clocks"-composite molecular markers thought to capture an individual's biological age-have been traditionally developed through bulk-level analyses of mixed cells and tissues. However, recent evidence highlights the importance of gaining single-cell-level insights into the aging process. Microglia are key immune cells in the brain shown to adapt functionally in aging and disease. Recent studies have generated single-cell RNA-sequencing (scRNA-seq) datasets that transcriptionally profile microglia during aging and development. Leveraging such datasets in humans and mice, we develop and compare computational approaches for generating transcriptome-wide summaries from microglia to establish robust and applicable aging clocks. Our results reveal that unsupervised, frequency-based summarization approaches, which encode distributions of cells across molecular subtypes, strike a balance in accuracy, interpretability, and computational efficiency. Notably, our computationally derived microglia markers achieve strong accuracy in predicting chronological age across three diverse single-cell datasets, suggesting that microglia exhibit characteristic changes in gene expression during aging and development that can be computationally summarized to create robust markers of biological aging. We further extrapolate and demonstrate the applicability of single-cell-based microglia clocks to readily available bulk RNA-seq data with an environmental input (early life stress), indicating the potential for broad utility of our models across genomic modalities and for testing hypotheses about how environmental inputs affect brain age. Such single-cell-derived markers can yield insights into the determinants of brain aging, ultimately promoting interventions that beneficially modulate health and disease trajectories.