Abstract
Dystrophin-deficient zebrafish larvae are a small, genetically tractable vertebrate model of Duchenne muscular dystrophy well suited for early stage therapeutic development. However, current approaches for evaluating their impaired mobility, a physiologically relevant therapeutic target, are characterized by low resolution and high variability. To address this, we used high speed videography and deep learning-based markerless motion capture to develop linked-segment models of larval escape response (ER) swimming. Kinematic models provided repeatable, high precision estimates of larval ER performance. Effect sizes for ER peak instantaneous acceleration and speed, final displacement, and ER distance were 2 to 3.5 standard deviations less for dystrophin-deficient mutants vs. wild-types. Further analysis revealed that mutants swam slower because of a reduction in their tail stroke frequency with little change in tail stroke amplitude. Kinematic variables were highly predictive of the dystrophic phenotype with ≤ 3% of larvae misclassified by random forest and support vector machine models. Tail kinematics also performed as well as in vitro assessments of tail muscle contractility in classifying larvae as mutants or wild-type, suggesting that ER kinematics could serve as a non-lethal proxy for direct measurements of muscle function. In summary, ER kinematics can be used as precise, physiologically relevant, non-lethal biomarkers of the dystrophic phenotype. The open-source approach described here may have applications not only for studies of skeletal muscle disease but for other disciplines that use larval mobility as an experimental outcome.