The central nervous system (CNS) generates complex motor commands for locomotion by combining motor modules, represented as muscle synergies. Existing research suggests that muscle synergies are either innate or established early in life, but development and skill acquisition likely necessitate their refinement or alteration. This study aims to clarify how these locomotor synergies evolve during development and training. The biomechanics of running differ significantly between children and adults, and adult performance running requires specific motor training. Running, therefore, serves as an ideal model to explore the adaptability of neural constraints during development and learning. The study uses electromyographic (EMG) data from multiple muscle groups to characterize muscle synergies using Non-negative Matrix Factorization (NMF). This algorithm identifies time-invariant multi-muscle activation profiles (muscle synergies) that are scaled by time-varying coefficients. The sum of these scaled synergies approximates observed multi-muscle activation patterns. The correspondence between NMF-derived muscle synergies and spinal interneuronal network co-activation of multiple muscles supports their use in this study. Surface EMGs (15 right-sided lower-limb muscles) were recorded during running at self-selected speeds for various subject groups.

Prior research suggests that motor modules, or primitives, are either inborn or determined early in development, exhibiting significant invariance across the lifespan and even species. However, other studies indicate that motor development and skill acquisition may require the acquisition of new modules or modification of existing ones. Understanding how early motor modules are updated by the CNS to meet developmental and learning needs is a crucial question in neuroscience. Previous research on running biomechanics highlights differences between children and adults, and the role of training in optimizing adult running economy. The use of muscle synergy analysis, particularly NMF, has been established in characterizing motor modules and their relation to neural network activity.

A combined cross-sectional and longitudinal design was employed, encompassing five groups: preschoolers (Presch; age 3–6), sedentary adults (Sedent; no prior training), novice adult runners (Novice), experienced runners (Exp), and elite runners (Elite). EMGs were recorded from 15 right-sided lower-limb muscles during running at self-selected speeds. Sedentary and novice adults underwent multiple recording sessions (0, 2 months for Sedent; 0, 3, 6 months for Novice) at the same speed. Preschoolers ran overground, while adults ran on a treadmill. Kinematic data were also collected using motion capture. Ground reaction forces were recorded to assess energetic efficiency (energy loss per kg of body mass). EMGs underwent preprocessing (high-pass, rectification, low-pass filtering, integration, spike removal, normalization). NMF was used to extract muscle synergies, and the number of synergies required for ~80% EMG reconstruction was determined. K-means clustering identified representative synergy vectors, and their sparseness (number of active muscle components) was quantified. Synergy similarity was assessed using scalar products. Muscle synergy fractionation (splitting of a synergy into multiple units) and merging (combination of synergies) were modeled as linear combinations of synergy vectors. A Merging Index (MI) was created to quantify the percentage of synergies explainable by merging sedentary adult synergies. Energetic efficiency was correlated with specific synergy merging patterns to identify combinations associated with enhanced or reduced efficiency. Temporal patterns of synergy activation coefficients (C(t)) were analyzed to explore the mechanisms of synergy merging. Statistical analyses (t-tests, Mann-Whitney U tests, ANOVA, Kruskal-Wallis tests, repeated measures ANOVA, Friedman's test, Pearson's correlation) were used to compare groups and assess correlations.

The study revealed significant developmental and training-related plasticity of muscle synergies for running. During development (Presch to Sedent), muscle synergies fractionated into sparser units with fewer muscles. Three preschooler synergies, primarily involving extensors, fractionated into six sedentary synergies, showing a consistent pattern of tibialis anterior (TA)-only fractionation and tensor fasciae latae (TFL)/gluteus maximus (GLUT) fractionation. This was further validated by analysis of individual subject pairs. In contrast, adult running training led to a decrease in synergy dimensionality and an increase in synergy sparseness. The Merging Index showed a gradual increase from sedentary to elite runners, indicating merging of pre-training synergies. Five synergy merging combinations were significantly associated with running efficiency (p=0.011-0.049, 2-tailed Mann-Whitney). Three combinations (SO-7+11, 5+6+8, 5+6+12, where SO denotes Sedent0 cluster number) were associated with higher efficiency, and their prevalence increased from sedentary to elite runners. Two combinations (SO-3+12, 4+5+7) correlated with reduced efficiency, with SO-3+12 being more prevalent in sedentary runners than in more experienced runners. Analysis of synergy merging patterns revealed that the presence of efficiency-reducing patterns (R1, R2) resulted in low efficiency regardless of the presence of efficiency-enhancing patterns (E1, E2). High running efficiency was strongly associated with the presence of E1 or E2, and the absence of R1 or R2. Analysis of synergy activation coefficients suggested that synergy merging may result from reassigning multiple original synergy-encoding networks to be driven by one of the original oscillators. The results indicated that the CNS uses pre-existing synergies to learn new skills rather than creating new patterns de novo.

The findings support the plasticity of muscle synergies in response to development and training. Developmental fractionation adapts early synergies to changing neuro-musculoskeletal properties during growth. Training-induced merging refines motor commands, reducing degrees of freedom and focusing on efficient movement patterns. Specific merging patterns are linked to enhanced (e.g., co-activation of extensors for propulsion, arm-leg swing synchronization) or reduced (e.g., co-contraction of ankle plantar/dorsiflexors, inefficient arm swing) efficiency. The study suggests that muscle synergy analysis can help identify inefficient running strategies not easily detected visually. The low efficiency in experienced runners, possibly due to prevalence of a specific merging pattern, highlights potential bottlenecks in training. While running speed correlated with synergy changes, it is not the sole determinant. Longitudinal data from novice runners show that synergy changes occur even at constant speed, emphasizing the role of training itself. The proposed neural mechanism for synergy merging involves reassignment of synergy-encoding networks to be driven by a single oscillator, potentially guided by motor cortical inputs and reinforced by feedback. Developmental fractionation and training-induced merging are distinct processes, with merging often involving combinations different from those formed during fractionation. The study highlights that developmental factors influence adult motor learning by shaping the available building blocks for subsequent training adaptations.

This study demonstrates the plasticity of muscle synergies through fractionation and merging during running development and training. Fractionation adapts early synergies to developmental changes, while merging refines synergies during training, leading to improved efficiency. Specific merging combinations are linked to efficient or inefficient running, and the CNS appears to modify existing synergies rather than creating new ones during skill acquisition. Future research should investigate the interaction between developmental fractionation and training-induced merging, exploring how these processes are affected by factors such as early physical activities, injuries, and nutrition.

The study used a cross-sectional and longitudinal design with limited sample sizes, particularly for some of the adult subgroups. The energetic efficiency was estimated based on vertical ground reaction force, which may not fully capture all aspects of running energetics. Further, the self-selected speeds for running may have varied somewhat within groups and across sessions in some subjects, potentially confounding the effects of development and training. The study focused on running, and the findings may not generalize to other motor tasks.