Muscle synergies are largely unaffected in individuals with diabetes who do not have diabetic neuropathy

Abstract

Background:

Impaired neuromuscular function in individuals with diabetes can adversely affect gait kinematics, kinetics, and electromyography, potentially increasing the risk of serious complications such as plantar ulcers and amputations. However, it remains unclear whether these changes are associated with alterations in muscle synergies. This study aims to examine muscle synergies in individuals with diabetes.

Methods:

Surface electromyography recordings were obtained from seven lower extremity muscles (vastus lateralis, rectus femoris, biceps femoris, semitendinosus, lateral gastrocnemius, soleus, and tibialis anterior) during 20 trials of barefoot walking. Eleven individuals with type 2 diabetes without diabetic neuropathy and ten age-matched controls were recruited. Variations in synergy complexity were assessed by the number of synergies needed to account for >90 % of the total variance in the electromyography data, total variance accounted for by one synergy, and total variance accounted for by four synergies. Synergy weights and activations for a four-synergy solution were compared using cosine similarity. An electromyography co-contraction index was computed for agonist and antagonist pairs of muscles.

Findings:

Those with diabetes did not significantly differ from controls in the number of synergies, total variance accounted for by one synergy, total variance accounted for by four synergies, or synergy composition. However, they demonstrated higher levels of variability in synergy composition similarity.

Interpretation:

These results indicate that, at the group level, individuals with diabetes without neuropathy employ largely similar motor control strategies as their healthy counterparts while walking, and previously reported variations in gait biomechanics in this population may be attributed to peripheral neuromuscular dysfunction.

1. Introduction

Diabetes mellitus and the metabolic derangements associated with chronic hyperglycemia cause a variety of detrimental effects on neuromuscular function. These ramifications include degenerative changes in nerve fibers, compromised neural conduction abilities, as well as muscular atrophy and diminished strength, which culminate in impaired motor control and abnormal gait biomechanics among those with diabetes (Bril, 2014; Orlando et al., 2016; Tesfaye et al., 2010). Previous electromyography (EMG) studies demonstrated altered muscle activity and timing during gait in individuals with diabetes. For example, during heel strike, the peak EMG activity of the tibialis anterior is delayed in those with diabetes (Abboud et al., 2000; Sacco and Amadio, 2003), although this muscle remains active for a longer duration compared to healthy controls (Kwon et al., 2003; Savelberg et al., 2010). Peak amplitude of the lateral gastrocnemius has been delayed (Akashi et al., 2008; Gomes et al., 2011; Sacco et al., 2010; Sawacha et al., 2012), whereas the soleus is activated earlier and remains active for a longer duration than healthy controls (Kwon et al., 2003). The peak EMG activity of the vastus lateralis is delayed (Akashi et al., 2008; Kwon et al., 2003; Sacco et al., 2010; Sacco and Amadio, 2003), whereas vastus medialis activity lasts significantly longer than healthy controls (Savelberg et al., 2010). Peak rectus femoris activity occurs earlier upon heel strike (Sawacha et al., 2012) while greater co-contractions of the tibialis anterior and triceps surae muscles have been observed (Kwon et al., 2003). Although abnormal muscle activity and EMG are often reported in those with diabetic neuropathy, emerging evidence suggests that similar changes can occur in individuals with diabetes who do not exhibit clinically significant peripheral neuropathy (Savelberg et al., 2010; Sawacha et al., 2012). These findings hint at the possibility that pathological changes in the central nervous system’s control of movement may be responsible for the abnormal muscle activity observed in those with diabetes.

Muscle synergy analysis identifies groups of lower-dimensional weighted muscle activations that are often recruited together to perform functional tasks (Bernstein, 1967; Sherrington, 1892; Tresch and Jarc, 2009). By decomposing the EMG signal, muscle synergy analysis allows a more holistic evaluation of grouped muscle activities while also serving as a means to quantify the effectiveness of motor coordination in healthy and neurological populations (Shuman et al., 2017). Previous studies have examined muscle synergies in a variety of neurological disorders, including stroke (Cheung et al., 2009; Cheung et al., 2012; Clark et al., 2010; Roh et al., 2013), Parkinson’s disease (Falaki et al., 2016; Rodriguez et al., 2013; Roemmich et al., 2014), and cerebral palsy (Hashiguchi et al., 2018; Li et al., 2013; Shuman et al., 2019; Steele et al., 2015; Tang et al., 2015). Impaired motor control is often demonstrated by a reduction in the complexity and number of muscle synergies. (Cheung et al., 2012; Falaki et al., 2016; Steele et al., 2015; Tang et al., 2015). For example, children with cerebral palsy demonstrate fewer muscle synergies when performing dynamic tasks (Shuman et al., 2019; Steele et al., 2015). A reduced number of synergies, known as “merging,” may indicate poor motor coordination and occurs when one synergy module compensates for the dysfunction of another and meets the functional demands of a task (Cheung et al., 2012; Clark et al., 2010).

Several changes in gait kinematics, kinetics, and EMG have been reported in individuals with diabetes (Dingwell and Cavanagh, 2001; Fernando et al., 2013; Hazari et al., 2016; Menz et al., 2004; Rao et al., 2010). These variations may lead to elevated plantar pressures and loading during walking and facilitate the development of plantar ulcers. However, it remains unknown whether these changes are accompanied by abnormalities in muscle synergies among individuals with diabetes. Muscle synergy analysis may be useful for developing effective measures of disease progression in people with diabetes or have utility in identifying patients at higher risk of foot ulceration or amputation.

The purpose of this study is to investigate whether individuals with diabetes demonstrate altered synergies, including changes in the number of synergies, synergy complexity, weights, or activations. To differentiate the effects of diabetic neuropathy from other complications associated with diabetes mellitus, the current study tested individuals with diabetes without neuropathy. We hypothesize that individuals with diabetes without neuropathy exhibit altered synergies compared to healthy controls.

2. Methods

2.1. Participants

Eleven individuals with type 2 diabetes mellitus (DM), without a history of diabetic neuropathy (age, 53 ± 11 yrs.; body mass, 87 ± 16 kg; body height, 159 ± 7 cm; leg length 82 ± 6 cm HbA1C, 7.37 ± 1.80; duration of disease 75 ± 79 months) and ten non-diabetic controls (CON), (age, 52 ± 14 yrs.; body mass, 76 ± 15 kg; body height, 166 ± 9 cm; leg length 87 ± 7 cm) participated in the study. Participants were excluded if they had a history or presence of diabetes-related neuropathy, diabetic foot ulcers, foot or lower limb amputation, severe lower extremity musculoskeletal injury, cerebral injury, neurological disorders such as Parkinson’s disease, and major lower extremity surgeries. All procedures were approved by the Institutional Review Board at the University of Kentucky (Reference# 62242). All participants signed an informed consent form approved by the institutional review board.

2.2. Experimental process

Prior to data collection, each participant completed a questionnaire containing a brief medical history (time since diagnosis, HbA1c levels, family history of diabetes, smoking, drinking, depression or mental illness, exercise, history of cardiovascular disease, history of high blood pressure). In addition, a 5.07 / 10 g Semmes-Weinstein monofilament test was conducted to confirm the absence of peripheral neuropathy. Body height, body mass, leg length (anterior superior iliac spine to lateral malleolus), and shank length (lateral femoral epicondyle to lateral malleolus) were measured for each individual. A balance recovery test was used to determine the dominant leg (Hoffman et al., 1998). Surface EMGs (Delsys Inc., Boston, Massachusetts) were collected at 2500 Hz from seven lower extremity muscles: vastus lateralis (VL), rectus femoris (RF), biceps femoris (BF), semitendinosus (ST), lateral gastrocnemius (LGS), soleus (SO), and tibialis anterior (TA). EMG preparation and electrode placement were carried out on the dominant limb ipsilaterally following SENIAM guidelines (Hermens et al., 2000). Kinematics and kinetics data were collected synchronously using a three-dimensional motion capture system (Motion Analysis Corp., Rohnert Park, CA) and three force platforms (two Bertec force plates – Bertec Corp., Columbus, OH – and one AMTI force plate – Advanced Mechanical Technology, Inc., Watertown, MA) at 250 and 2500 Hz, respectively. Each participant completed 20 successful trials of barefoot walking, while ensuring clean foot contact with the force plates. Gait events were identified using a vertical ground reaction force threshold of 20 N for both heel strike (first instant above threshold) and toe-off (first instant below threshold) (Zeni Jr et al., 2008). In each trial, three consecutive gait cycles were collected, and the second gait cycle was used for further EMG analysis to avoid any variations associated with gait initiation or termination. Individuals with diabetes typically walk slower than healthy controls, with an average gait speed of 0.89–0.9 m/s (Chiles et al., 2014; Rao et al., 2006). To control for its potential confounding effect, walking speed was kept consistent for all participants at a fixed pace of 0.89–0.9 m/s using a wireless TCi timing system (Brower Timing Systems, Draper, UT).

2.3. Electromyography data

A linear envelope was computed for each muscle and trial using the following steps: the EMG data were demeaned to remove offset, bandpass filtered (20–400 Hz) using a 4th order Butterworth bandpass filter to remove motion artifacts and high-frequency noise, rectified, and finally low-pass filtered with a 4th order Butterworth filter (20 Hz cutoff frequency) to obtain a linear envelope (De Luca et al., 2010; Mello et al., 2007). The use of a 4th order Butterworth filter is well-established in previous muscle synergy analysis literature (Banks et al., 2017; Clark et al., 2010; Rabbi et al., 2020; Shuman et al., 2017). For each trial, the linear envelope of the second gait cycle—defined from heel strike to heel strike of the dominant leg—was extracted. The resulting envelope was down-sampled to 100 Hz, yielding 101 points per gait cycle, to reduce computational demands while preserving the temporal resolution necessary for synergy analysis, in line with previous studies (Banks et al., 2017; Chvatal and Ting, 2013; Shuman et al., 2017). The gait cycles from all 20 trials were concatenated and amplitude-normalized to the peak activity of each respective muscle across all trials (Chvatal and Ting, 2013; Clark et al., 2010; Shuman et al., 2017).

2.4. Synergy analysis

Synergies were calculated from the processed EMG data using non-negative matrix factorization (NNMF) in MATLAB (Statistics and Machine Learning Toolbox, MathWorks Inc., Natick, MA) (Chvatal and Ting, 2013; Lee and Seung, 1999; Rabbi et al., 2020; Ting and Chvatal, 2010). NNMF assumes that a muscle activation pattern (EMG_mxt) is composed of a linear combination of a few muscle synergies (W_mxn) that are each recruited by a synergy activation coefficient (C_nxt) such that EMG_mxt = W_mxn ×C_nxt + error, where m is the number of muscles, t is the number of time points, and n is the number of synergies. Synergy compositions (W) were fixed, and each synergy was multiplied by a time-varying scalar activation coefficient (C). In NNMF, random initial guesses were generated and then replicated until minimal error was obtained. The following NNMF settings were used: 50 replicates, 1000 maximum iterations, a 1 × 10⁻⁴ termination tolerance for the relative change in the elements of W and C, and a 1 × 10⁻⁶ termination tolerance for the change in the size of the residual. The amplitude of each muscle activity vector was scaled to have a unit variance to assure equal weighting before extracting synergies, (Roh et al., 2013; Steele et al., 2013; Ting and Chvatal, 2010). After synergy extraction, data were reconstructed by multiplying the W and C vectors (EMG_recon = W × C), and the goodness of fit of the reconstructed data by n number of synergies was measured by the variance accounted for (VAF) (Chvatal and Ting, 2013; Torres-Oviedo et al., 2006; Zar, 1999). We used two measures to evaluate synergy complexity: (1) number of synergies required for overall VAF greater than 90 % and VAF of each muscle greater than 75 % (Nsyn), and (2) total variance accounted for by one single synergy (tVAF1) and total variance accounted for by four synergies (tVAF4). The number of synergies, as an independent measure, has been frequently used to assess synergies in both healthy and neuro-compromised populations (Clark et al., 2010; Ivanenko et al., 2004; Routson et al., 2013). Also, it has been shown that the total variance accounted for by one single synergy (tVAF1) serves as a key indicator of the synergy complexity, especially in clinical populations (Schwartz et al., 2016; Shuman et al., 2018). Accordingly, Nsyn, tVAF1, and tVAF4 were used to evaluate whether DM differed from CON In terms of synergy complexity.

Additionally, we assessed whether synergy weights or activations varied between DM and CON. For 70 % of the participants in CON, at least four synergies were responsible for over 90 % of the variation in the EMG data. Accordingly, synergy weights and activations for a four-synergy solution were calculated for both DM and CON, respectively. Synergies were sorted functionally, based on their structural similarity, muscle composition and synergy activation profiles (Torres-Oviedo and Ting, 2007). We then created a control archetype by averaging the synergy weights and activations of the 4-synergy solution across all CON. Each participant’s individual synergies in both CON and DM were matched to their closest counterparts in the archetype. To evaluate how closely an individual’s synergies resemble those in the control archetype, we calculated cosine similarity (CS) scores between each participant’s synergy weights and activations and their corresponding counterparts in the control archetype. Cosine similarity (CS) measures the degree of similarity between two non-zero vectors on a scale of 0 to 1, with higher values indicating more similar synergies (0 = no similarity, 1 = perfect similarity). For example, the cosine similarity for each synergy weight (W_n), was calculated as the dot product of the vectors divided by the product of their Euclidean norms, as shown below:

$${\text{CS}}_{\text{n}}=\frac{W_{\text{n}}.W_{\text{n Arch}}}{‖W_{\text{n}}‖‖W_{\text{n Arch}}‖}$$

2.5. EMG Co-contraction index

The resulting linear envelopes were used to assess the level of co-contraction between the agonist-antagonist pairs of muscles including ankle plantar flexors and dorsiflexors (TA-LGS and TA-SOL), as well as knee extensors and flexors (VL-BF and RF-SE). These muscle pairs were selected based on biomechanical and anatomical factors such as their relative anatomical proximity, their roles in major joint functions during gait, and prior literature (da Fonseca et al., 2004; Kwon et al., 2003; Strazza et al., 2017). A co-contraction index (CCI) was obtained for each pair of muscles by overlaying the linear envelopes, calculating the area of overlap, and then dividing the area of overlap by the number of data points (Frost et al., 1997). The resultant CCI was used to compare muscle co-contraction between the DM and CON throughout all 20 gait cycles.

2.6. Statistical analyses

Statistical analyses were conducted in MATLAB R2019a. The significance level was set at 0.05. Descriptive statistics, including mean and standard deviation were computed for each dependent variable. Independent samples t-tests were employed to compare synergy complexity metrics (Nsyn, tVAF1, and tVAF4) and muscle co-contraction index between DM and CON. To assess synergy compositions, cosine similarity scores between individual synergy vectors and a control archetype were calculated for both groups. A Wilcoxon rank sum test was used to compare cosine similarity scores between DM and CON groups for each synergy.

After extracting the optimal number of synergies (4) for both groups, greater inter-subject variability in muscle contributions within each synergy was observed in DM compared to CON. Specifically, the relative weighting of individual muscles within each synergy were more heterogeneous in DM, while they were more uniform in CON. To quantify these differences, post-hoc F-tests for equality of variances were conducted to compare the variability in each muscle’s relative contribution within each synergy vector between DM and CON. Bonferroni correction was applied to adjust for multiple comparisons.

3. Results

3.1. Synergy complexity

3.1.1. Nsyn, tVAF1 and tVAF4

Seventy percent of individuals in the CON group had four or more synergies (6/10 four synergies and 1/10 five synergies), while the remaining 30 % had three synergies (tVAF>90 % and VAF muscle>75 %). DM demonstrated a similar trend, with 64 % (7/11) of individuals in DM having four synergies and the remaining 36 % (4/11) having three synergies. DM did not differ significantly from CON in Nsyn (p = 0.518).

The total variance accounted for by one single synergy (tVAF1) was 69.28 % ± 9.43 % and 72.56 % ± 8.07 % for CON and DM respectively, which was not statistically significant (p = 0.400). The tVAF4 for a four-synergy solution averaged 93.29 % ± 2.42 % and 93.54 % ± 2.14 % for CON and DM respectively (p = 0.803) (Fig. 1).

Fig. 1. –

tVAF for a given number of synergies in DM and CON (left) and Number of synergies for tVAF>90 % and VAF muscle>75 % (right).

3.1.2. Synergy composition

Fig. 2 illustrates the average synergy weights and activations for both DM and CON, respectively. Although DM generally showed reduced similarity to the control archetype, Wilcoxon rank sum tests revealed no statistically significant differences in similarity scores for a four-synergy solution between DM and CON. This was true for both synergy weights (S1: p = 0.072; S2: p = 0.459; S3: p = 0.275; S4: p = 0.148) and activations (S1: p = 0.915; S2: p = 0.418; S3: p = 0.084; S4: p = 0.169).

Fig. 2. -.

Average synergy weights and activations for a four-synergy solution in CON (A) and DM (B). The percentage in the top-right corner of each synergy weight graph represents that synergy’s contribution to the overall variability of the reconstructed data. Since each synergy (S_i) accounts for a portion of the total reconstructed data variability, its percentage contribution to the overall variability of the reconstructed data was calculated as ∑(W_i c_i) / ∑(WC)*100), where W_i and c_i correspond to synergy weights and activation coefficients for synergy i, respectively, and WC denotes the full reconstructed dataset.

The average cosine similarity of synergy weight vectors for CON ranged from 0.955 to 0.986 across the four synergies, while for DM it ranged from 0.878 to 0.960. Specifically, the average cosine similarity in CON were 0.986, 0.975, 0.955, and 0.975 for W1 (LGS, SOL), W2 (ST, BF), W3 (VL, RF), and W4 (TA) respectively, compared to 0.960, 0.902, 0.878, and 0.928 for DM. Similarly, the average cosine similarity for synergy activations in CON ranged from 0.788 to 0.903, while in DM it ranged from 0.747 to 0.902. The average cosine similarity in CON were 0.903, 0.788, 0.818, and 0.824 for C1 (LGS, SOL), C2 (ST, BF), C3 (VL, RF), and C4 (TA) respectively, compared to 0.902, 0.747, 0.755, and 0.779 for DM.

Post hoc F-tests for equality of variances revealed significantly greater variability in relative muscle weights to synergies: S2 (VL, p < 0.001; RF, p = 0.001), S3 (TA, p < 0.001), and S4 (BF, p < 0.001) in DM compared to CON. For example, the relative contribution of TA, SOL, and RF to S3 synergy weights was 0.054 ± 0.063, 0.101 ± 0.066 and 0.919 ± 0.135 in CON but 0.151 ± 0.278, 0.158 ± 0.150 and 0.835 ± 0.313 in DM.

3.1.3. Muscle co-contraction

DM demonstrated a slightly higher CCI compared to CON, particularly for the TA-SOL, RF-ST, and VL-BF muscle pairs. For example, for the VL-BF pair, CCI was 7.347 ± 3.865 in CON while 9.640 ± 4.955 in DM. Similarly, for the RF-ST pair, CCI was 9.379 ± 5.393 in CON but 10.642 ± 8.015 in DM. Nevertheless, these differences were not statistically significant: TA-LGS, (p = 0.761), TA-SOL (p = 0.328), VL-BF pair (p = 0.069), and RF-ST (p = 0.439), (See Fig. 3.).

Fig. 3.

EMG muscle co-contraction index (CCI).

4. Discussion

The major findings of this study indicate that people with diabetes without neuropathy use similar movement control strategies known as muscle synergies when compared to healthy control participants during walking. To the best of our knowledge, this is the first study to evaluate muscle synergies in individuals with diabetes.

Previous studies have documented alterations in gait kinematics, kinetics, and electromyography among individuals with diabetes (Abboud et al., 2000; Akashi et al., 2008; Alam et al., 2017; Fernando et al., 2013; Gomes et al., 2011; Kwon et al., 2003; Sacco et al., 2009; Sacco et al., 2010; Sacco and Amadio, 2003; Savelberg et al., 2010; Sawacha et al., 2012). These include changes in muscle activation timing and increased co-contractions in agonist-antagonist pairs of muscles, which may indicate poor motor coordination during gait and possibly lead to an increased risk of falls, injuries, and long-term hospitalizations in this population (Abboud et al., 2000; Akashi et al., 2008; Gomes et al., 2011; Hoveizavi et al., 2023; Kwon et al., 2003; Sacco et al., 2010; Sacco and Amadio, 2003; Savelberg et al., 2010; Sawacha et al., 2012). Several mechanisms may contribute to impaired neuromuscular function in people with diabetes. First, glycation of myosin may impair both the structural and functional characteristics of skeletal muscles (Ramamurthy et al., 2001). Such alterations are often accompanied by changes in skeletal muscle metabolism, leading to reductions in muscle mass, strength, power, and quality (Orlando et al., 2016). More often, diabetic peripheral neuropathy is assumed to account for the pathogenesis of those impairments. However, an increasing body of evidence suggests that hyperglycemia directly impacts the ability of the muscles to generate force (Kalyani et al., 2015). Second, diabetes mellitus is associated with a variety of neuropathies, ranging from focal to diffuse types, which often lead to sensory, motor, and autonomic dys-functions. Damage to small fibers has been observed early in the development of diabetic sensorimotor neuropathy and may even occur at the prediabetes stage (Bril, 2014). The cumulative effect of these mechanisms likely contributes to the subtle alterations in muscle contributions within each synergy in individuals with diabetes, highlighting the complex nature of diabetic neuromuscular dysfunction.

In this study, muscle synergy analysis revealed similar synergies, at the group level, in individuals with diabetes without neuropathy when compared to healthy control participants, in terms of both synergy complexity and composition. The number of synergies for a tVAF cutoff greater than 90 % is commonly used as a measure of synergy complexity, with a reduced number of synergies indicating decreased complexity. Our results revealed no significant differences between DM and CON in the number of synergies for tVAF greater than 90 % and VAF muscle greater than 75 %. Additionally, more rigorous local criteria for synergies accounting for greater than 75 % of VAF in each muscle were also added to ensure that critical features of the EMG data set were reproduced (Chvatal and Ting, 2013). Likewise, tVAF1 and tVAF for n number of synergies have been previously used as measures of synergy complexity, with higher values indicating a lower level of complexity and impaired synergies being characterized by significantly higher tVAF and tVAF1 values (Steele et al., 2015; Tang et al., 2015). In this study, no significant differences were found between DM and CON with respect to either tVAF1 or tVAF4. Our results suggest that synergy complexity (as measured by Nsyn, tVAF1, and tVAF4) may not be significantly affected by diabetes – at least before the onset of diabetic neuropathy – in contrast to other clinical populations (Shuman et al., 2019; Steele et al., 2015; Tang et al., 2015).

Our results on synergy composition in a four-synergy solution were in agreement with previously reported findings using somewhat similar muscle combinations (Magrath et al., 2022; Neptune et al., 2009; Tang et al., 2015): S1 demonstrated ankle plantar flexor activity (LGS and SOL) primarily contributing to forward propulsion during late stance. S2 represented hamstrings activity (ST, BF), decelerating the forward movement of the limb during terminal swing and preparing the foot for initial contact. S3 involved activations in the quadriceps (VL, RF), stabilizing the hip and knee for weight acceptance, and S4 reflected ankle dorsiflexor activity (TA), providing foot clearance during early swing.

Although there were no statistically significant differences in overall synergy structure between DM and CON, our analysis revealed increased variability in synergy composition among individuals with diabetes. This variability was particularly evident in cosine similarity scores, where DM showed a wider range of similarity scores compared to CON. For instance, the average cosine similarity scores for synergy weights ranged from 0.955 to 0.986 in CON, but 0.878 to 0.960 in DM. For synergy activations, the scores ranged from 0.788 to 0.903 in CON, and from 0.747 to 0.902 in DM. In addition, post-hoc analysis revealed greater variability in relative muscle weightings across synergies in DM. This increased heterogeneity in muscle contributions may reflect subtle neuromuscular alterations in individuals with diabetes, potentially indicative of early-stage muscle deficits, weakness, or subclinical neuropathic processes occurring prior to the manifestation of clinically detectable diabetic peripheral neuropathy. A tendency towards increased muscle co-contraction index was observed, particularly for the VL-BF muscle pair, which was indeed reflected in our S2 synergy module.

Though not statistically significant at the group-level, these trends suggest subtle changes in motor control strategies among individuals with diabetes, potentially indicating early shifts in muscle coordination in this population. The observed variability could be influenced by factors beyond neuropathy, such as metabolic alterations or subtle biomechanical adaptations. These changes may reflect impaired motor control and coordination, which could potentially increase the risk of falling in people with diabetes. The heterogeneity in synergies across those with diabetes highlights the complex nature of motor control adaptations in this population. Although our analysis revealed only subtle variations, it lays the groundwork for exploring the potential of muscle synergy analysis as a tool for assessing neuromuscular changes in individuals with diabetes. Further research, especially in those with diabetic neuropathy, is needed to determine whether alterations in muscle synergies can serve as an early indicator of disease progression or identify those at higher risk of complications such as falls or foot ulceration.

4.1. Limitations

The quality of EMG data could be affected by factors such as electrode-skin motion artifact, muscle crosstalk, and inter-subject variability. In addition, the methodology used in the muscle synergy analysis may have a significant impact on its results. For example, the number and choice of muscles impacts the maximum number of synergies that can be identified, as well as the relative structure of each synergy weight (Steele et al., 2013). The EMG pre-processing, including the low pass filtering cutoff frequency and amplitude normalization, may affect synergy complexity including tVAF1, tVAF4, and Nsyn (Shuman et al., 2017). Additionally, EMG may contain residual variability due to sensorimotor noise or other neural mechanisms that are not accounted for by muscle synergies (e.g. short-latency reflex responses) (Ting, 2007). Variations in the body height and mass of the two cohorts (DM and CON) may present potential confounding factors that could affect the outcome of this study. Finally, synergies may be influenced by gait speed (Escalona et al., 2021). In this study, both groups walked at a slightly slower pace of 0.89–0.9 m/s. It remains unknown whether more significant group differences would emerge at higher gait speeds.

5. Conclusion

This study demonstrates that individuals with diabetes without neuropathy use similar muscle synergies while walking compared to their non-diabetic counterparts. The observed similarity in muscle synergies suggests that significant adaptations in movement control may not yet be present at this stage of the disease. Further investigation is needed to understand the intricate interaction between peripheral and central mechanisms in motor control, especially as diabetes progresses.