A Cross-Condition Single-Cell Atlas Defines Skeletal Muscle Adaptation and Its Predictive Limits

Skeletal muscle adapts to diverse stresses through coordinated changes across myofibres and cellular niches, yet these responses are usually studied separately. We assembled a cross-species discovery atlas of 501,940 cells and nuclei from six public resources, comprising 2863 condition-time-state combinations defined at the declared analysis unit level, and evaluated it with two independent validation datasets. Injury produced the largest observed compositional displacement, followed by mechanical overload, whereas acute human endurance exercise and denervation showed smaller occupancy shifts despite substantial cell-type-specific molecular responses. Genetic lineage data identified distinct newly fused myonuclear programmes; the overloaded Myomaker-knockout versus sham comparison remained limited by loading, genetic background and batch differences. Spatial analyses identified local enrichment of a terminal Schwann cell expression module around neuromuscular junction regions, and paired human datasets provided design-sensitive translational support. Across 20 fixed model specifications with training-only predictor preprocessing, the baseline-plus-context control model showed small average improvements over composition-only and intervention permutation controls, with heterogeneous performance across study holdouts. Here, we show that an evidence-graded atlas can characterise observed adaptive remodelling alongside study-dependent baseline structure while defining the limits of prospective prediction based on coarse composition.