Shared muscle synergies between running and physical performance tests Authors: Hiroki Saito, Ayu Yamano, Nanae Suzuki, Kazuya Matsushita, Hikaru Yokoyama, Atsushi Sasaki, Tatsuya Takahashi, Sakiko Ito, Kimitaka Nakazawa Published: 2025 ABSTRACT BackgroundUnderstanding whether commonly used physical performance test (PPT)-related tasks replicate the neuromechanical characteristics of running is essential for informing return-to-running (RTR) decisions. PPTs are frequently used in rehabilitation, yet their ability to reflect the neuromuscular demands of running has not been fully established. This study therefore investigated the extent to which muscle synergies are shared between running and various PPT-related tasks to evaluate the neuromechanical replicability of these tasks.MethodsTen healthy male participants were recruited, performing running and nine PPT-related tasks including walking, single-leg hops (30, 60 and 100% of their maximum distance), single-leg squats, and various balance tasks. The activities in sixteen lower limb and trunk muscles were recorded. The data were analyzed using non-negative matrix factorization to explore shared muscle synergies between running and PPT-related tasks. The percentages of shared synergies and temporal patterns were compared between running and each PPT-related task.ResultAll PPT-related tasks exhibited shared muscle synergies with running. However, walking (75% [40%-100%]), single leg hops with 30% of maximum distance (60% [20%-100%]), and step-up-down (63% [0%-100%]) demonstrated significantly higher percentages of shared synergies compared to other tasks. However, significant differences in temporal patterns were observed between running and all PPT-related tasks.ConclusionWalking, SLH30, and step-up-down tasks share a substantial proportion of muscle synergy patterns with running and appear to more closely reproduce the coordination strategies used during running. These tasks may therefore be promising candidates for future studies aimed at evaluating the predictability of safe return to running. Background Returning to running (RTR) represents a key milestone in the rehabilitation process following sports-related injuries, including anterior cruciate ligament (ACL) rupture, hamstring strain, and meniscal injury [1, 2]. While the course of rehabilitation and the appropriate timing of RTR differ across injurie types, severities, and the duration of sport inactivity, the phase of RTR bridges early rehabilitation—focused on restoring knee range of motion and everyday function—into later stages that involve higher-intensity activities such as jumping, cutting, and other sport-specific drills [3]. The timing of RTR carries significant clinical importance: resuming running too soon can markedly elevate the risk of re-injury or graft failure[4], whereas excessive delays in return to running may lead to detraining and hinder recovery, and can also reduce motivation and psychological readiness for sport [5, 6]. Clinicians must therefore carefully balance safety and progression when determining an athlete’s readiness to run [1, 7]. Physical performance tests (PPTs) are used clinically to assess the successful execution of tasks including walking, single-leg hops, single-leg squats (SLS), and various balance tasks, aiding clinicians in determining readiness to RTR [1, 7]. PPTs are designed to replicate some of the physical requirements and neuromechanical control strategies inherent in running behaviors [7, 8]. Clinicians qualitatively assess whether the lower limbs have sufficient capacity to cope with the demands of running activities [7, 8]. However, in the context of RTR, PPTs are infrequently and inconsistently applied, and there is no clear consensus on which tests, if any, best predict safe running resumption [7]. The infrequent and variable use of PPTs suggests that some athletes may return to running without thorough evaluation of functional readiness, whereas others may have their running unnecessarily delayed. This gap is problematic because inadequate functional recovery (e.g. persistent strength deficits or poor movement mechanics) at the time of RTR may contribute to abnormal running gait, elevated joint loading, or compensatory strategies that increase the risk of subsequent injury. A key unresolved question is whether passing commonly used PPTs actually indicates that an athlete’s neuromuscular system is prepared for the demands of running. In other words, do these tests possess neuromechanical validity as proxies for running, and are they appropriate candidates for future prognostic research? If a test is meant to evaluate readiness to run, it should reasonably challenge the same muscle groups and coordination patterns that are required during running. Thus, before the predictive value of PPTs can be rigorously established, it is essential to understand to what extent these tasks replicate the neuromechanical characteristics of running. Identifying tasks that more closely reproduce running-like neuromuscular coordination would provide a rational basis for selecting which PPTs should be prioritized in future studies that directly evaluate the predictability of safe RTR. Muscle synergy analysis offers a novel way to assess the neuromuscular underpinnings of movement [9]. The central nervous system is thought to simplify motor control by recruiting muscles in groups or "modules" (synergies) rather than controlling each muscle in isolation. In healthy runners, 4–5 muscle synergies are sufficient to reconstruct the muscle activation patterns of a full gait cycle, aligning with the phases of running (weight acceptance, propulsion, early swing, late swing)​ [10–12]. Such synergies represent coordinated recruitment of multiple muscles (e.g., a “weight acceptance” synergy typically involves synchronous activation of the quadriceps, gastrocnemius, and gluteals to stabilize and absorb impact at landing). If a particular PPT elicits muscle synergy patterns that closely resemble those of actual running, it would support that test’s neuromechanical validity for RTR decisions. Conversely, if the synergies used in a test differ markedly from running synergies, the test might be a poor surrogate for running readiness. This information is invaluable for clinicians, as it helps in selecting more appropriate PPTs that truly emulate running when making return-to-running decisions, thereby strengthening the evidence basis for safe and effective RTR assessments.​ Accordingly, the purpose of this study was to investigate the degree of muscle synergy overlap between running and a spectrum of PPTs used in RTR assessments. This study was designed as an exploratory step to identify which tasks exhibit stronger neuromechanical similarity to running and therefore may warrant further evaluation in prognostic research. We hypothesized that the proportion and functional types of muscle synergies shared with running would differ across tasks. To comprehensively assess synergy similarity, our task selection encompassed both commonly used RTR tests in literature [7] and exploratory movements (e.g., a heel raise, and a “diver” balance maneuver), ensuring that the comparisons spanned a broad range of functional activities. Methods Study design and participants This study employed a cross-sectional design and was conducted at a local university. Ten healthy males, with a mean (SD) age of 21 (± 0.3) years old, were recruited from a local university. The mean height and weight of the participants were 173.8 (± 3.8) cm and 67.5 (± 8.0) kg, respectively. All participants reported no history of lower-limb injury within the previous 12 months. Each participant provided written informed consent to participate in the study. The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the University of Tokyo. Procedures Participants were asked to freely perform running at a jogging pace and PPT-related tasks (Fig. 1). The majority of PPT-related tasks were selected based on tasks commonly used in RTR following ACL injuries [7]. This included normal walking [13, 14], and the single leg hops (SLH) tests, where participants executed a forward jump [15, 16]. Although the standard procedure for SLH is asking subjects to jump forward 100% (SLH100) of their maximum distance, this may not be appropriate in the RTR phase of rehabilitation, especially for patients who have undergone an ACLR. Thus, we exploratorily added SLH at 30% (SLH30) and 60% (SLH60) of their maximum distance. A single leg squat (SLS) was also carried out with approximately 45 degrees of knee flexion, and the step-up-down task involved a forward step up onto a 10 cm high box and return to the floor [14]. The Y-Balance test protocol involves the subject standing with one foot on a stance plate at the starting line, then reaching with the opposite leg in three directions: anterior, posteromedial, and posterolateral [17]. Furthermore, since most PPT-related tasks commonly focused on weight acceptance with the activations of quadriceps and gluteus muscle groups [18, 19], while running also involves the plantar flexors for push-off and knee flexors for swing deceleration [10–12]. Thus, we added the single-legged heel-rise (Hraise) to bias toward the activations of plantar flexors [20] and the diver, which is commonly used in rehabilitation for hamstring injuries, for the activations of hamstrings [21]. Diver describes an exercise where participants were asked to stand on their right foot with their left knee bent at 90° and the left foot raised [21]. They were instructed to flex their hip while stretching their arms forward and to attempt maximal hip extension of the raised leg while keeping the pelvis level [21]. Of note, running at a jogging pace and walking at a comfortable speed showed average velocities of 2.1 ± 0.28 m/s and 1.2 ± 0.12 m/s, respectively. Each PPT-related task was repeated five times, and the order of the tasks was randomly assigned.Fig. 1Running and nine PPT-related tasks. Participants independently performed running at a jogging pace and a walk. In the single leg hops (SLH) tests, participants executed a forward jump at 30% (SLH30), 60% (SLH60), and 100% (SLH100) of their maximum distance. A single leg squat (SLS) was carried out with approximately 45 degrees of knee flexion. The Y-Balance test involved standing on one leg on a stance plate while reaching with the opposite leg in the anterior, posteromedial, and posterolateral directions. Hraise refers to the heel raise task, while step-up-down denotes a forward step up onto a 10 cm height box. Diver describes an exercise where participants were asked to stand on their right foot with their left knee bent at 90° and the left foot raised. They were instructed to flex their hip while stretching their arms forward and to attempt maximal hip extension of the raised leg while keeping the pelvis level Running and nine PPT-related tasks. Participants independently performed running at a jogging pace and a walk. In the single leg hops (SLH) tests, participants executed a forward jump at 30% (SLH30), 60% (SLH60), and 100% (SLH100) of their maximum distance. A single leg squat (SLS) was carried out with approximately 45 degrees of knee flexion. The Y-Balance test involved standing on one leg on a stance plate while reaching with the opposite leg in the anterior, posteromedial, and posterolateral directions. Hraise refers to the heel raise task, while step-up-down denotes a forward step up onto a 10 cm height box. Diver describes an exercise where participants were asked to stand on their right foot with their left knee bent at 90° and the left foot raised. They were instructed to flex their hip while stretching their arms forward and to attempt maximal hip extension of the raised leg while keeping the pelvis level Unilateral surface EMG data were recorded from 16 lower limb and trunk muscle groups: rectus abdominis (RA) (3 cm lateral to umbilicus) [22], oblique externus (OE) (15 cm lateral to umbilicus) [23], erector spinae at L1 (ESL1) (3 cm lateral to the L1 spinous process) [22], gluteus maximus (GM), gluteus medius (Gmed), biceps femoris (long head, BF), semitendinosus (ST), tensor fasciae latae (TFL), adductor longus (ADD), rectus femoris (RF), vastus medialis (VM), vastus lateralis (VL), tibialis anterior (TA), gastrocnemius medialis (MG), soleus (SOL) and peroneus longus (POL). The EMG sensor placements in lower limb were based on SENIAM (surface EMG for a non-invasive assessment of muscles) [24]. A wireless EMG system (Trigno Wireless System; DELSYS, Boston, MA, USA) was used to record EMG activity. Each electrode had an inter-electrode spacing of 10 mm. The EMG signals were band-pass filtered (20–450 Hz), amplified (with a 300-gain preamplifier), and sampled at 2000 Hz using an analog-to-digital converter (Power lab/16SP, AD Instruments, Australia). Marker coordinate data were collected at 120 Hz using an eight-camera motion capture system (Vicon, Centennial, CO) with a 25-marker set. This set incorporated markers for the head, arms, trunk, pelvis, thighs, shanks, and feet, based on the Vicon Plug-in-Gait model. Marker coordinate data were interpolated using cubic spline interpolation to remove gaps in the data and filtered with a low-pass third-order Butterworth filter at 20 Hz. This data was then combined with subject-specific anthropometric data to create an eight-segment whole-body model. The kinematic profiles, calculated by the Vicon Plug-in-Gait model, were used to define the start and end of each trial for each task. Raw EMG signals were high-pass filtered at 30 Hz to remove motion artifacts, and then demeaned. The signals were then full-wave-rectified and low-pass filtered at 10 Hz, using a fourth-order Butterworth filter. The smoothed EMG envelopes were time-interpolated to generate 200 time points between the start and end points for each trial so that the EMG data of each trial contributed to the extracted muscle synergies equally. We created single-task EMG matrices for each task for each participant (that is, the matrix was composed of 16 muscles × 1000 timepoints (the no. of repetitions/cycles (5) × 200 samples)) to extract the muscle synergies for each task. Each EMG from each muscle was normalized to the maximum amplitude across all tasks. To extract muscle synergies, non-negative matrix factorization (NMF) was applied to the single-task EMG matrix. NMF has previously been described as a linear decomposition technique [25] according to Eq. (1):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M=W\cdot C+e$$\end{document} Where M (m × t matrix, where m is the number of muscles, t is the number of samples [i.e., spatiotemporal profiles of muscle activity]) is a linear combination of muscle synergies, W (m × n matrix, where n is the number of muscle weighting component), C (n × t matrix, representing temporal components), and e is the residual error matrix. We applied NMF to extract possible n values from 1 to 16 for each dataset. To estimate the optimal number of muscle synergies, the variance accounted for (VAF) by the reconstructed EMG (M) was calculated at each iteration [26]. VAF was defined as 100 × the square of the uncentered Pearson's correlation coefficient [26, 27]. Considering the local minima inherent in NMF, each synergy extraction was repeated 50 times, and the VAF was calculated at each iteration. Iterations with the highest VAF were maintained [12, 28]. VAFs > 0.9 were used to identify the optimal number of synergies commonly used in the literature [12, 28]. To extract the number of shared and running-specific, test-specific muscle synergies, we used a modified version of the NMF algorithm based on the previous studies [29, 30] that simultaneously extracts motor modules that are shared across running and each task and those that are specific to each task from a data matrix containing EMG from both conditions. We defined a shared synergy as one activated in both tasks, with temporal pattern components showing non-zero coefficients in each. To identify task-specific muscle synergies, the coefficients, C, corresponding to running were set to zero (i.e., test-specific synergies), and those corresponding to each task were are set to zero (i.e., run-specific synergies). Detailed descriptions of this method are described elsewhere [29]. Briefly, asin independent synergy extraction, the number of shared, runningspecific and task-specific synergies for each participant was determined as the minimum number of total muscle synergies required to exceed a VAF of 90%. We defined the percentage of shared muscle synergies between running and each task as the ratio of the number of shared modules to the number of total motor modules across the two tasks. This modification is thought to improve the accuracy of shared and task-specific muscle synergies by minimizing the possibility that the numbers of synergies are underestimated when synergies are extracted from the EMG of each task independently, and by allowing comparison of the similarity of synergies between tasks to identify shared and task-specific synergies [31]. We identified representative shared, running-specific and test-specific synergies across participants using hierarchical clustering analysis (Ward's method, Euclidian distance) of muscle weighting components [32]. To determine the optimal number of clusters, we computed the gap statistic [33], which measures the compactness of the clustering achieved against those in reference data sets without any obvious clustering similar to a previous study [34]. Reference data sets (N = 500) were initially generated by sampling uniformly from within the bounds of the original muscle-synergy set; each of them was then clustered by the hierarchical cluster, at 2–20 clusters. The optimal number of clusters was then the smallest number, h, such that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{Gap}\left(\mathrm{h}\right)\ge\text{Gap }\left(\text{h }+1\right) -\text{sd }\left(\text{h }+1\right)$$\end{document}where Gap(k) represents the gap statistic at h clusters, and sd(h) signifies the standard deviation of the clustering compactness within the reference data sets [33]. We defined shared, running-specific and test-specific synergy clusters as having each three types of synergies from ≥ 1/2 of synergies within a cluster. If the cluster was not contributed by any types of synergies, we defined it as “none-classified”. Statistical analyses We compared the percentage of shared muscle synergies between tasks. The values were compared using the Friedman test, which is a non-parametric method for multiple comparisons of paired samples, as a normal distribution was not observed in the data (tested using the Shapiro–Wilk test). When the Friedman test showed significant effects, multiple comparison post-hoc analyses were performed using the Wilcoxon signed-rank test. Temporal pattern components were compared between running and PPTs using statistical parametric mapping (SPM) (spm1d v0.4.7 for MATLAB, Institute of Neurology, London, UK) [35]. Since each single-task EMG matrix contained 5 repetitions/cycles, the extracted temporal pattern components of running and PPTs were converted into an averaged repetition/cycles for each participant before comparing temporal pattern components using SPM analysis. The p values obtained from were corrected using the false discovery rate (FDR) correction for multiple comparisons [36]. The significance level for all tests was set at p < 0.05. When there was a significant difference between the groups, effect sizes (ES) were calculated using Cohen’s d [37]. We recruited ten participants without an a priori power analysis, thus, we instead conducted a sensitivity analysis in G*Power, which indicated that an effect size of 0.87 would be necessary to obtain a power of 80% at an α of 0.05. Results Table 1 summarizes the VAF values and the number of synergies for running and each PPT-related task. Across tasks, VAF values were consistently high (≥ 0.91), indicating robust reconstruction of the EMG signals. The number of extracted synergies ranged from 3 to 5 across tasks.Table 1VAF values and synergy numbers for running and PPT tasksVAF valuesNumber of synergiesRunning0.92 (± 0.01)3.9 (± 0.31)PPT-related tasksWalk0.92 (± 0.01)3.8 (± 0.63)SLH300.92 (± 0.01)4.3 (± 0.67)SLH600.91 (± 0.01)4.6 (± 0.51)SLH1000.91 (± 0.01)5.1 (± 0.56)SLS0.92 (± 0.01)2.4 (± 0.51)Ybalance0.92 (± 0.01)3.6 (± 0.84)Hraise0.92 (± 0.01)2.3 (± 0.67)Step-up-down0.91 (± 0.01)4.5 (± 0.97)Diver0.92 (± 0.01)3.4 (± 0.69)VAF values represent mean (± SD) variance accounted for the reconstructed EMG signals. Synergy numbers indicate the minimum modules required to reach VAF ≥ 0.90 Table 1VAF values and synergy numbers for running and PPT tasksVAF valuesNumber of synergiesRunning0.92 (± 0.01)3.9 (± 0.31)PPT-related tasksWalk0.92 (± 0.01)3.8 (± 0.63)SLH300.92 (± 0.01)4.3 (± 0.67)SLH600.91 (± 0.01)4.6 (± 0.51)SLH1000.91 (± 0.01)5.1 (± 0.56)SLS0.92 (± 0.01)2.4 (± 0.51)Ybalance0.92 (± 0.01)3.6 (± 0.84)Hraise0.92 (± 0.01)2.3 (± 0.67)Step-up-down0.91 (± 0.01)4.5 (± 0.97)Diver0.92 (± 0.01)3.4 (± 0.69)VAF values represent mean (± SD) variance accounted for the reconstructed EMG signals. Synergy numbers indicate the minimum modules required to reach VAF ≥ 0.90 VAF values and synergy numbers for running and PPT tasks VAF values represent mean (± SD) variance accounted for the reconstructed EMG signals. Synergy numbers indicate the minimum modules required to reach VAF ≥ 0.90 Figure 2 depicts the percentage of muscle synergies shared between running and each PPT-related task. The percentages differed significantly among the nine tasks (Friedman test, p = 0.000079). The primary finding was that walking, SLH30, and step-up-down demonstrated the highest percentages of shared synergies with running. Statistically significant differences were observed between the following pairings: walking vs. SLS (p = 0.0038, ES = 2.18), walking vs. Hraise (p = 0.0098, ES = 2.05), SLH30 vs. SLS (p = 0.0017, ES = 1.56), SLH30 vs. Hraise (p = 0.0039, ES = 1.47), SLS vs. step-up-down (p = 0.0072, ES = 1.31), and Hraise vs. step-up-down (p = 0.0176, ES = 1.25). The median percentages of shared synergies for each task were: walking (75% [40–100%]), SLH30 (60% [20–100%]), SLH60 (50% [28–75%]), SLH100 (50% [28–80%]), SLS (32% [16–50%]), Y-balance (40% [25–80%]), Hraise (22% [20–60%]), step-up-down (63% [0–100%]), and Diver (40% [20–80%]).Fig. 2The percent of shared muscle synergies between running and each PPT task. Median values are indicated as horizontal lines inside the boxes. The edges of the boxes represent the 25th and 75th percentiles. The percentage of shared synergies was significantly different among the nine PPTs (p = 0.000079 for both groups, Friedman test one-way ANOVA). Statistically significant differences were observed between the following pairings: Walk and SLS (p = 0.0038, ES = 2.18), Walk and Hraise (p = 0.0098, ES = 2.05), SLH30 and SLS (p = 0.0017, ES = 1.56), SLH30 and Hraise (p = 0.0039, ES = 1.47), SLS and step-up-down (p = 0.0072, ES = 1.31), and Hraise and stepup (p = 0.0176, ES = 1.25) The percent of shared muscle synergies between running and each PPT task. Median values are indicated as horizontal lines inside the boxes. The edges of the boxes represent the 25th and 75th percentiles. The percentage of shared synergies was significantly different among the nine PPTs (p = 0.000079 for both groups, Friedman test one-way ANOVA). Statistically significant differences were observed between the following pairings: Walk and SLS (p = 0.0038, ES = 2.18), Walk and Hraise (p = 0.0098, ES = 2.05), SLH30 and SLS (p = 0.0017, ES = 1.56), SLH30 and Hraise (p = 0.0039, ES = 1.47), SLS and step-up-down (p = 0.0072, ES = 1.31), and Hraise and stepup (p = 0.0176, ES = 1.25) Figure 3 shows representative shared, running-specific, and PPT-specific muscle synergies identified by cluster analysis (muscle weighting components (W) and temporal pattern components (C)). Running consisted of four synergies in total. W1 mainly represented hip (Gmax and Gmed) and knee extensors (RF and VM) for weight acceptance. W2 showed ankle extensors (GL, SOL, and PL). W3 represented trunk (RAS, OE, and ESL1), hip flexors (TFL), and ankle dorsiflexors (TA). Finally, W4 showed knee flexors (BF and ST) and ankle dorsiflexors (TA). These synergies were used as references to order the shared synergies in PPT-related tasks. All PPT-related tasks shared muscle synergies with running, although the number of shared, PPT-specific synergies and non-classified synergies differed between each task. For example, walking shared all four synergies with running, while there were no PPT-specific or non-classified synergies. On the other hand, SL30 shared two synergies (W2 and W3) with one non-classified synergy. The SPM analysis found significant differences in temporal pattern components between running and each PPT-related task (p < 0.05).Fig. 3Representative shared, running-specific, and test-specific muscle synergies identified by cluster analysis (muscle weighting components (W) and temporal pattern components (C)). The gray sections on the X-axis of C represent the temporal pattern components where a significant difference between running and PPT-related tasks was observed (p < 0.05). RA: rectus abdominis, OE: oblique externus, ESL1(erector spinae at L1), GM: gluteus maximus, Gmed: gluteus medius, BF: biceps femoris, ST: semitendinosus, TFL: tensor fasciae latae, ADD: adductor longus, RF: rectus femoris, VM: vastus medialis, VL: vastus lateralis, TA: tibialis anterior, MG: gastrocnemius medialis, SOL: Soleus (SOL) and POL: peroneus longus Representative shared, running-specific, and test-specific muscle synergies identified by cluster analysis (muscle weighting components (W) and temporal pattern components (C)). The gray sections on the X-axis of C represent the temporal pattern components where a significant difference between running and PPT-related tasks was observed (p < 0.05). RA: rectus abdominis, OE: oblique externus, ESL1(erector spinae at L1), GM: gluteus maximus, Gmed: gluteus medius, BF: biceps femoris, ST: semitendinosus, TFL: tensor fasciae latae, ADD: adductor longus, RF: rectus femoris, VM: vastus medialis, VL: vastus lateralis, TA: tibialis anterior, MG: gastrocnemius medialis, SOL: Soleus (SOL) and POL: peroneus longus Discussion In this study, we applied muscle synergy analysis to compare running with nine PPT-related tasks commonly used in RTR assessments. Our primary aim was to identify which tasks exhibit neuromechanical features that most closely resemble those of running and therefore may be promising candidates for future prognostic research. We found that all tested tasks shared at least some synergies with running, indicating that these activities recruit broadly comparable muscle group combinations. Notably, walking, SLH30, and the step-up-down task exhibited the greatest overlap in muscle synergy patterns with running. This pattern suggests that these tasks engage neuromuscular coordination strategies more closely aligned with those required for running. Taken together, our findings indicate that certain PPT-related tasks—particularly walking, low-level hopping, and step-up tasks—may demonstrate relatively greater neuromechanical similarity to running and may therefore be prioritized in future studies that directly evaluate the predictability of PPTs for safe return to running. Our approach was originally derived from foundational work in motor control showing that the CNS organizes movement through a limited set of modular “shared synergies,” which are flexibly recombined across different behaviors [38]. Although previous studies have investigated shared synergies between various motor tasks—such as walking and running [11], walking and cycling [39], and balance tasks [40]—no prior research has examined shared synergies between PPT-related tasks and running in the context of sports injuries and return-to-play decision-making. In the present study, we assumed that a high degree of shared synergies between running and PPT-related tasks would signify these tests as strong predictors of running behavior, due to similar neuromechanical characteristics, including the coordination of multiple muscles. Firstly, our results showed that the percentage of shared synergies (muscle weighting components) between running and walking was 75% [40%−100%], with all synergies between the two tasks being shared. This aligns with a previous study that suggested consistent muscle weighting components for walking and running, facilitating the transition from walking to running [11]. Similarly, we found that the SLH 30 and the step-up-down test shared a significant amount of muscle synergies with running, which is reasonable given that both activities share the same mechanical goal of propelling the body's center of mass forward [41]. This action requires dynamic balance with appropriate activations of lower and trunk muscles. The SLH test, particularly the SLH30 variant, exhibited a higher degree of shared muscle synergies with running compared to other non-jumping tests. Conversely, SLH60 and SLH100 demonstrated a relatively low number of shared synergies, as they also generated test-specific synergies beneficial for longer forward jumping. NMF also extracted temporal pattern components, signifying the activation profiles of muscle weighting components. Notably, there were significant differences in some temporal pattern components of muscle synergies between running and all PPT-related tasks. From a neuromechanical perspective, these temporal patterns reflect the specific timing and intensity of muscle activation, illustrating that even though muscle synergies may be similar, the way they are employed in different activities can differ significantly. This discrepancy could be attributed to the unique demands of each task. Even though similar muscle groups may be engaged (thus the shared synergies), the coordination, timing, and intensity of muscle activation might not perfectly align [42]. From a clinical perspective, these differences imply that PPT-related tasks cannot fully replicate the mechanical demands of running. Specifically, the intensity of temporal components of walking, SLH30, and step-up-down tasks that exhibited higher percentages of shared synergies seemed to be lower than those in running. Running necessitates the production of substantial strength, particularly to withstand impact forces, which can be two to three times one's body weight [43]. Consequently, PPT-related tasks may not replicate the same intensity of mechanical output produced by muscle synergies and thus, may underestimate running abilities. We also note that the present study characterized neuromuscular patterns at a single time point in healthy participants and did not investigate long‑term clinical outcomes. We therefore cannot determine whether using these tasks to guide RTR decisions improves rehabilitation efficiency, reduces time to return to running, or lowers re‑injury risk. These questions will require longitudinal studies in injured populations that link task‑based assessments with subsequent clinical outcomes. This study has several limitations. First, the sample size was small (n = 10) and consisted only of healthy young males, which limits the generalizability of the findings to other populations, such as females, older adults, or patients undergoing rehabilitation. Second, surface EMG was used, which may not fully capture the activity of deeper muscles. Deeper muscles may contribute to trunk and lower-limb coordination during locomotion, and their absence from the recordings may have led to an incomplete representation of the full neuromuscular control strategy. Third, running was assessed only at a jogging pace. Jogging was intentionally selected because it reflects the first phase of RTR in clinical practice and provides a safe and standardized condition for synergy analysis. Nevertheless, neuromuscular demands increase with speed, and future work should investigate whether task–running similarity changes under faster running conditions. Conclusion This study demonstrates that walking, SLH30, and step‑up‑down tasks share a substantial proportion of muscle synergy patterns with running. From a neuromechanical perspective, these tasks appear to more closely reproduce the coordination strategies used during running and may therefore be promising candidates when selecting PPTs for future studies that directly evaluate the predictability of safe return to running. However, given the small and homogeneous sample of healthy young males, these findings should be interpreted cautiously and not as definitive clinical recommendations; further research in larger and more diverse cohorts is needed to confirm the clinical utility of these tasks.