A network-based atlas of human skeletal muscle aging

Skeletal muscle metabolic and physical capacities are influenced by both genetics and load status and decline with age. Recent advances in sequencing have detailed cell types at unprecedented detail; yet these approaches do not scale to adequately model human muscle physiological heterogeneity. We produced a powerful resource for ageing studies, including consistent deep transcriptomic profiles of 1,675 human muscle biopsies (∼28,000 genes per profile) and multiple single-cell spatial transcriptomic technologies. We present several novel models of tissue ageing. Five Quantitative network models (QNMs), built using >40 trillion calculations and 930 human muscle transcriptomes, modelled aging and the influence of load status. Additional differential expression (DE) signatures for atrophy, hypertrophy and cardio-respiratory adaptation were integrated with single-cell RNAseq and cell-specific bulk profiles to reveal cell-enriched modules and the topology of human skeletal aging. Rapamycin transcriptomes from cultured muscle and endothelial cells, along with in vivo signatures for insulin resistance and sex, were integrated into these analyses. We show that >3,000 genes are DE with muscle age (equally up and down); that a novel pre-frailty signature in elderly subjects has a remarkably strong overlap with the response of healthy muscle during experimental atrophy and that the hypertrophy signature in elderly muscle, but not young muscle, opposes the age-regulated transcriptome. We report that non-responders for hypertrophy or gains in cardio-respiratory capacity have highly distinct genome-level response to exercise. QNM revealed cell-specific processes in endothelial cells and fibroblasts, including novel interactions between insulin sensitivity, age and senescence. From two hundred and eighty-six hub genes consistent in both young and old muscle network models, 27% had known roles in muscle biology, while of the top 50 hub genes (45% protein coding), 80% were newly linked to human muscle biology, including ARHGAP4, CEP131 and IFITM10 and many short- and long-noncoding RNAs. Many genes demonstrated extreme changes in topology in old muscle, such as the neddylation and aging linked gene, DCUN1D5. GeoMX-based spatial muscle fibre-type profiling (57 regions), along with Xenium (8 regions) and Merscope (54 regions) single-cell spatial technologies located key aging, frailty and load-responsive genes to individual cell types and provided novel insight into the location of autocrine/paracrine secreted factors such as GDNF, while IL6 was located to rare endothelial cells. A machine-learning model ranked the factors most associated with the topological changes with age. This prioritised network features over DE signatures, highlighting positive correlating edges to down-regulated genes during atrophy, genes up-regulated by Rapamycin and both positive and negative correlating insulin sensitivity features, along with gene hub status, best explained muscle ageing. Genome level modelling produced an independently validated transcriptomic 'age clock' and found it to be invariant to muscle load status in people >50y, while we revealed novel interactions between gene length and age. Release of an unprecedented level of consistently aligned genomic data, along with QNMs with >7,000 searchable modules, provides a powerful resource for the aging research communities.