Resistance Training Molecular Signatures Identify Pharmacodynamic Biomarker Candidates in Duchenne Muscular Dystrophy

Abstract Duchenne Muscular Dystrophy (DMD) is characterized by progressive skeletal muscle loss for which validated pharmacodynamic biomarkers remain a critical unmet need. Biomarker discovery in DMD is fundamentally constrained by small patient cohorts and the ethical limitations of repeated muscle biopsy. Here we report that molecular signatures confirmed across seven independent exercise physiology datasets, spanning three species, four mass spectrometry platforms, and multiple independent laboratories, provide a high-confidence prior for evaluating directionally consistent findings in DMD muscle biopsies. The stretch-sensor protein ANKRD2 (ankyrin repeat domain-containing protein 2) shows tri-context convergence across three independent human cohorts: resistance-training (RT) inducible in healthy young muscle (6-site phosphosignature), aging-blunted in long-term trained older adults (interaction log₂FC −0.611, p=0.025), and markedly depleted in DMD muscle biopsies (log₂FC = −4.54). This tri-context pattern positions ANKRD2 as a mechanistic anchor linking the exercise-adaptation deficit of aging to the structural-signaling collapse of dystrophy. Integrating this anchor with therapeutic window intersection scoring of 147 established muscle genes across DMD protein deficit and RT inducibility evidence, and with Heterogeneous Graph Transformer (HGT) graph neural network analysis of 194,541 gene nodes, we identify a tiered pharmacodynamic biomarker panel. Tier 1 candidates UCP3 and FBXO32 are each supported by RNA evidence from at least two independent DMD datasets, protein-level depletion in DMD biopsies, and RT inducibility in aging muscle; UCP3 is independently cross-validated by GNN neighborhood analysis. The panel spans four biological axes: sarcomere integrity, calcium-exercise signaling, proteostasis, and metabolic reprogramming. We propose that cross-cohort reproducibility in a related physiological domain constitutes a principled mechanism for confidence transfer to underpowered rare disease datasets, and that this framework, anchored by a mechanistically grounded protein with tri-context convergence, is directly applicable to proof-of-concept endpoint design in DMD therapy trials and broadly generalizable to other rare neuromuscular conditions.