The Lower Limb Muscle Co-activation Map during Human Locomotion: From Slow Walking to Running.

1. Introduction

Muscle coactivation is one of the strategies used by the central nervous system (CNS) to simplify movements and ensure adequate joint stiffness by regulating the time and amplitude of simultaneous activity of a pair or group of muscles [1,2,3,4]. Muscle coactivation is thought to maintain effector-level control (low dimensional), removing the need for individual muscle coordination control (high dimensional) [2].

Although muscle coactivation has been extensively studied during gait in both healthy and patients with a variety of motor disorders (for a review, see [2], only a few studies have looked at muscle coactivation during running [5,6,7,8,9]. Walking to running represents a critical transition time that corresponds to a change in CNS activation [10], leg geometry [11,12], joint compliance [13], and a robust mechanical energy transformation [14,15,16,17].

Most humans will voluntarily switch from walking to running at 6.8 and 7.5 Km/h [18,19,20,21]. At these higher speeds, running becomes less expensive than walking by utilizing a mass-spring mechanism that exchanges kinetic and potential energy in very different ways [15,17]. Most studies on muscle activation during running have concentrated on the activity of single muscles [22,23,24]. Running has been linked to increased activation of all lower limb muscles in general [5,25,26,27], despite the fact that some muscles (e.g., gluteus maximus) appear to play a larger role in running than in walking [28]. Only a few studies have looked at the activation of muscle pairs while running [9,29]. When compared to a single-joint muscle solution, these studies discovered that a longer duration of muscle coactivation between the rectus femoris and gastrocnemius during stance provided a better metabolic solution to multiple joint stability, implying that biarticular muscles redistribute mechanical power from proximal joints to distal joints, and ultimately to the ground. Moore et colleagues [9] discovered that coactivation in the distal and leg flexor muscles decreases with speed while running, implying that when both legs are off the ground, muscle coactivation must be reduced to allow for more leg propulsion both upwards and forwards.

A critical missing piece is whether the descending motor commands for running are set up to coactivate the lower limb muscles as a whole entity within a gross motor strategy. Taking previous evidence into account [9,29], it is possible to hypothesize that descending commands for running cause an increase in whole limb stiffness to stabilize the limb during ground contact on the one hand, and a decrease in limb stiffness to facilitate forward progression of the leg during air stepping on the other. This basic descending motor control would imply that flexor muscles are activated during air stepping, when whole-limb stiffness is reduced, and that muscle coactivation occurs via top-down (rostro-caudal) recruitment (from L3 to S2). This strategy would greatly simplify motor control at the low-dimensional effector level during shock absorption, allowing for motor control decentralization while relying on peripheral feedback via muscle-reflex to generate the required muscle coactivations and regulate each muscle's gain.

We recently proposed a novel approach to studying time-varying multi-muscle coactivation function (TMCf) [30,31] which is a good indicator of the CNS's overall strategy for modulating the simultaneous activation of many lower limb muscles during locomotion. This approach provides a new perspective on the spatiotemporal motor control of the lower limbs, emphasizing how all muscles of the lower limb are synchronously coactivated in an attempt to increase whole-limb stiffness, regardless of single-joint antagonist muscles or modular activation of a group of muscles [32].

The objectives of this study were to observe the differences in lower limb muscles coactivation strategies during locomotion across speeds from walking to running. We hypothesized that the behavior of lower limb muscles in terms of coactivation parameters, may differ across the speeds, particularly between walking and running, and that the coactivation parameters may vary based on the muscular functions (flexion or extensor functions), and across the rostro-caudal recruitment map.

2. Materials and Methods

2.1. Subjects

Nineteen healthy runners were recruited (9 men, 10 women; mean age, 40.95±7.96 years; mean weight, 66.47±14.60 kg). Each runner has declared of running for at least 5 years and that runs at least 3 times per week for at least 5 kilometers per training session.

None of the runners had any known diseases that might affect their regular gait or running pattern. The study was authorized by the local ethics committee (N. 0078009/2021) after all participants gave written informed consent and the study's design met with the Declaration of Helsinki.

2.2. Experimental procedure

Each runner was asked to walk at thirteen various speeds, ranging from 0.8 km/h to 6.8 km/h, and run at five different speeds, ranging from 7.3 km/h to 9.3 km/h, with increase steps of 0.5 km/h, on a treadmill. We chose 7.3 km/h as the transition speed because it is consistent with earlier researches [18,20,21] in which this transition was determined considering metabolic energy cost of locomotion, as well as the human capacity for purposeful gait modulation and the importance of physiologic and metabolic demands.

Before the recording session, the participants practiced for a few minutes to become acclimated to the task. To avoid fatigue, the trials were separated by 1-minute rest periods. Each trial was performed for 30 seconds before moving on to the next speed that was chosen at random.

2.3. Data acquisition

A bipolar 16-channel wireless acquisition device (Mini Wave System; Cometa, Italy) was used to record all the surface myoelectric activity at a sampling rate of 2 kHz. After skin preparation, twelve Ag/AgCl pre-gelled electrodes (Kendall ARBO, inter-electrode distance: 2 cm) were placed over each participant's right side lower limb muscles over the following muscles: gluteus medius (GM), rectus femoris (RF), vastus lateralis (VL), vastus medialis (VM), tensor fascia latae (TFL), semitendinosus (ST), biceps femoris (BF), tibialis anterior (TA), gastrocnemius medialis (GasM), gastrocnemius lateralis (GasL), soleus (S), and peroneus longus (P) (Merletti and Cerone, 2020; Merletti and Muceli, 2019), following the European Recommendations for Surface Electromyography (Hermens et al., 2000), the Atlas of Muscle Innervation Zones [33], and best practices [34,35,36].

A stereo-photogrammetric motion analysis system with optoelectronic technology (SMART-DX 6000 system: BTS, Italy) and with a sampling rate of 340Hz was employed to collect the kinematics data. Eight infrared cameras were used to record five passive spherical markers, covered with aluminum powder, positioned above the sacrum and bilaterally on the anterior superior iliac spines, heel, metatarsal head, and lateral malleoli (Davis et al., 1991).

A video camera (BTS Vixta; BTS, Italy) was used to record each task at a frame rate of 25-30 frames per second and a video resolution of 640x480 pixels.

All the data collected were synchronized.

2.4. Data analysis

3D reconstruction software (SMART Tracker and SMART Analyzer: BTS, Italy) and MATLAB (R2019b 9.7; MathWorks, USA) were used to process the sEMG and kinematics data.

2.4.1. Cycle definition and temporal normalization

The heel strike (HS) and toe-off (TO) events were determined in this study along the antero-posterior trajectories through the maximum point of the heel and the minimum point of the metatarsal, respectively. The gait and run cycle, were defined as the time between two consecutive HS of the same leg.

Ten cycles for each gait speed level for each runner were analysed and a polynomial approach was used to time-normalize the electromyographic and kinematic data on 201 samples [31,37,38,39].

2.4.2. Global, flexor, extensor and rostro-caudal coactivation of lower limb muscles

The raw sEMG signals were visually reviewed to remove any artifact-containing cycles. They were then band-pass filtered with a zero-lag fifth-order Butterworth (20–450Hz) [40,41] to keep only the signal of interest and then full wave rectified and low-pass filtered with a zero-lag fifth-order Butterworth (10 Hz) [42]. The elaborated sEMG signals of each muscle were amplitude-normalized (0-100%) for each runner in relation to the mean of their individual three largest peaks values detected throughout all cycles (Burden, 2010; Fiori et al., 2020).

The TMCf was used to calculate the simultaneous activation of the lower-limb muscles based on the processed sEMG signals [3,30,31,37,38,39,43]. The full-wave-rectified, low-pass-filtered, and 0–100% amplitude normalized sEMG signals were used as inputs to this sigmoid-weighted time-dependent function for the inclusion of multiple muscles during walking and running. This function's values ranged from 0 to 100%, and they were calculated as follows:

$$TMCf\left(d\left(i\right),i\right)=\left(1-\frac{1}{1+e^{-a\left(d\left(i\right)-b\right)}}\right).\frac{{({{\displaystyle \sum }}_{m=1}^{M}EM{G}_m\left(i\right)/M)}^2}{ma{x}_{m=1\dots M}\left[EM{G}_m\left(i\right)\right]}$$

(1)

where M is the number of muscles considered, $EM{G}_m\left(i\right)$ is the sEMG sample value of the $m-th$ muscle at instant $i$, $a$ and $b$ are constants equal to 12 and 0.5 respectively (Ranavolo et al., 2015), and $d\left(i\right)$ is the mean of the differences between each pair among the 12 $EM{G}_m\left(i\right)$ values at instant $i$:

$$d\left(i\right)=\left(\frac{{{\displaystyle \sum }}_{m=1}^{M-1}{{\displaystyle \sum }}_{n=m+1}^M\left|EM{G}_m\left(i\right)-EM{G}_n\left(i\right)\right|}{Lx\left(M!/\left(2!\left(M-2\right)!\right)\right)}\right)$$

(2)

$M!/\left(2!\left(M-2\right)!\right)$ is the total number of possible differences between each pair of $EM{G}_m\left(i\right)$.

$TMCf\left(d\left(i\right),i\right)$ has the following properties: inverse relationship with the mean of the differences d(i), i.e., values close to the mean activation of the $m\left(i\right)$ muscle sample values considered when $d\left(i\right)$ is close to 0, and values close to 0 when $d\left(i\right)$ is close to 1. The smaller the differences in muscle activations are, and the closer the $d\left(i\right)$ values are to 0, the closer the sigmoid-coefficient values are to 1, the closer the $TMCf\left(d\left(i\right),i\right)$ value close is to its mean value. In contrast, the greater the differences in muscle activations and the higher $d\left(i\right)$ and the lower the sigmoid coefficient, lowering the $TMCf\left(d\left(i\right),i\right)$ values. Data over individual strides was calculated for each runner and speed, and then averaged across cycles.

All the acquired muscles were inserted in the calculation of the TMCf to assess global coactivation (TMCf_glob). The coactivation of extensor (TMCf_ext), flexor (TMCf_flex) muscles separately, and according to the rostro-caudal organization (TMCf_L3; TMCf_L4; TMCf_L5; TMCf_S1; TMCf_S2) [44,45,46,47] was assessed using subgroups of muscles (see Table 1). Muscles were considered as flexors or extensors based on their concentric function in the sagittal plane [47]. The biarticular muscles were considered as flexors or extensors based on their proximal function [48,49].

2.4.3. Coactivation Parameters

Within the gait cycle, the following parameters were calculated for each map, runner and speed:

i) the synthetic coactivation index (CI_glob; CI_ext; CI_flex; CI_L3; CI_L4; CI_L5; CI_S1; CI_S2), which is calculated as the mean value of the TMCf and represents the average of the coactivation level [% coactivation]; ii) the maximum value of the TMCf (Max_glob; Max_ext; Maxflex; Max_L3; Max_L4; Max_L5; Max_S1; Max_S2) [% coactivation].

iii) the full width at half maximum (FWHM_glob; FWHM_ext; FWHM_flex; FWHM_L3; FWHM_L4; FWHM_L5; FWHM_S1; FWHM_S2) of the coactivation, which reflects the sum of the time durations, within the gait cycle, during which the TMCf curve is higher than its half maximum value [% gait cycle].

iv) the center of activity (CoA_glob; CoA_ext; CoA_flex; CoA_L3; CoA_L4; CoA_L5; CoA_S1; CoA_S2), which is calculated with circular statistics and plotted in polar coordinates to show where the most coactivation is concentrated within the walk and run cycles and, in this study, also to show the instant of coactivation onset on the rostro-caudal maps [% gait cycle] [31,50]:

$$\left\{\begin{array}{c}A={\displaystyle {\displaystyle \sum }_{i=1}^{201}}\left(\cos {\theta }_t\times EM{G}_i\right)\\ B={\displaystyle {\displaystyle \sum }_{i=1}^{201}}\left(\sin {\theta }_t\times EM{G}_i\right)\\ CoA={\tan }^{-1}\left(\frac{B}{A}\right)\end{array}\right.$$

(3)

v) the coefficient of multiple correlation (CMC), which measures the overall waveform similarity of a group of curves (the closer to 1 the CMC is, the more similar the curves are) [51,52,53]. CMC was calculated according to the following formula:

$$CMC=\sqrt{1-\frac{\left(1/(\mathrm{T}\left(\mathrm{N}-1\right))){{\displaystyle \sum }}_1^N{{\displaystyle \sum }}_1^T{\left(y_{it}-{\overline{y}}_t\right)}^2\right.}{\left(1/\left(\mathrm{TN}-1\right)){{\displaystyle \sum }}_1^N{{\displaystyle \sum }}_1^T{\left(y_{it}-\overline{y}\right)}^2\right.}}$$

(4)

$\mathrm{T}=200$ (number of points within the curve), $N$ is the number of curves, $y_{it}$ is the value at the $t-th$ point in the $i-th$ curve, ${\overline{y}}_t$ is the average of the two curves at point $t$, and ${\overline{y}}_t$ is the grand mean of all $y_{it}$.

Data over individual strides was calculated for each runner and speed, and then averaged across cycles. Within (CMC_Wt_glob, CMC_Wt_ext, CMC_Wt_flex, CMC_Wt_L3, CMC_Wt_L4, CMC_Wt_L5, CMC_Wt_S1, CMC_Wt_S2) and between (CMC_Bt_glob, CMC_Bt_ext, CMC_Bt_flex, CMC_Bt_L3, CMC_Bt_L4, CMC_Bt_L5, CMC_Bt_S1, CMC_Bt_S2) runners, the CMC was calculated.

2.4.4. Cross-correlation

The shape similarity overall the coactivation maps was evaluated using bidimensional normalized cross-correlation [31,37,54]. In detail, the central value of cross-correlation function $R\left(u,v\right)$, with an amplitude ranging from 0 to 1, was used to measured shape similarity between the global coactivation map and the extensor (R_G-E) and flexor (R_G-F) coactivation maps, as well as the shape similarity between the global coactivation map and the rostro-caudal coactivation maps (R_G-L3, R_G-L4, R_G-L5, R_G-S1, R_G-S2).

2.4.5. Center of Mass Displacement and Spatiotemporal parameters

The whole-body CoM for each speed and runner was calculated using the “reconstructed pelvis method” [55,56,57], i.e., the geometric center of the triangle formed by the markers over the two anterior superior iliac spines and the sacrum, which is the pelvic center. The displacement in vertical (CoM_y) and medio-lateral (CoM_z) directions was then calculated from the COM as the difference between the maximum and minimum values in their respective directions.

The following spatiotemporal parameters were determined for each speed and runner: i) Toe-off event (TO_e) [% gait cycle]; ii) stride length [cm]; iii) stride frequency [Hz]; iv) foot lift [cm] [37,56,58,59]; with the latter calculated as the maximum elevation along the vertical direction (y coordinate) of the centroid formed by the heel, metatarsal head, and lateral malleoli.

2.4.6. Statistical analysis

Repeated measurement ANOVA was used to determine the significance of differences in coactivation parameters computed from global, flexor, extensor, and rostro-caudal maps, as well as CoM and spatiotemporal parameters, with gait speed as a between-group factor (18 levels: from 0.8 to 9.3 Km/h), followed by a Bonferroni's pot-hoc analysis to determine whether there were significant differences between speeds.

An unpaired two-sample t test was used to compare the shape similarity of the global vs extensor coactivation maps and the global vs flexor coactivation maps. A univariate ANOVA was performed with coactivation map shape similarity as a between-group factor (5 levels: global vs each rostro-caudal coactivation map) and Bonferroni's post-hoc analysis was performed to assess the significant differences between each shape similarity.

The Watson-Williams test for circular data was applied to CoAs calculated from global, flexor, and extensor coactivation maps, with gait speed as the between-group factor, followed by a Bonferroni's pot-hoc analysis to see if there were any significant differences between the speeds [37,39,50,60]. Whereas the Harrison-Kanji test for circular data was applied to CoAs calculated from rostro-caudal maps, with gait speed and spinal level as between and within-group factors, a Bonferroni's pot-hoc analysis was used to determine whether there were significant differences between spinal levels and speeds [61].

A partial correlation analysis was performed, excluding the effects of gait speed, between each calculated global, flexor-extensor, rostro-caudal coactivation map, and each CoM and spatiotemporal parameter. Strong correlations were defined as significant correlation coefficients > 0.69 [62].

The significance level was set to 0.05, and all analyses were carried out in MATLAB (8.3.0.532, MathWorks, USA).

3. Results

3.1. Global, flexor, extensor and rostro-caudal coactivation maps and parameters

Figure 1 shows the three-dimensional global map of lower limb muscle coactivation from slow walking to running.

The map was created using the average TMCf_glob curves (black line) of 19 runners calculated on whole gait cycle, from 0% to 100% (x-axis), for each speed of walking and running performed, from 0.8 to 9.3 km/h (y-axis), and with amplitudes ranging from 0% to 100% of coactivation (z-axis). The TO_e (black dots) and FWHM_glob (the area underlying the TMCf_glob, red and grey for run and walk respectively) for each speed as a mean between all runners are also shown on the map.

Figure 2 shows the average and standard deviation of 19 runners' coactivation parameters (CI_glob, Max_glob, FWHM_glob, CMC_Wt_glob and CMC_Bt_glob) from slow walking to running. A significant main effect of gait speed was found on CI_glob, on the Max_glob, FWHM_glob, CMC_Wt_glob (see Table 2). The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values of CIglob, Max_glob, CMC_Wt_glob and CMC_Bt_glob were significantly higher, whereas the value of FWHM_glob was significantly lower (see Figure 2 and Table 2).

Figure 3 shows the average and the standard deviation of 19 runners' CoA_glob from slow walking to running. A significant effect of gait speed was found on CoA_glob (See Table 2). The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values of CoA_glob was significantly lower (see Figure 2, and Table 2).

Figure 4 shows the extensor (left side) and flexor (right side) maps of lower limb muscle coactivation from slow walking to running, with the averages TMCf_ext and TMCf_flex curves (black lines), TO_e (black dots), and FWHM_ext and FWHM_flex (the area underlying the TMCf_ext and TMCf_flex red and grey for run and walk respectively) of all 19 runners. A significant effect of speed was found on extensor and flexor parameters: CI_ext and CI_flex, on the Max_ext and Max_flex, FWHM_ext and FWHM_flex, CMC_Wt_ext, and CMC_Wt_flex (see Table 2). The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values of CI_ext, Max_ext, CMC_Bt_ext were significantly higher, whereas the value of CMC_Bt_flex was significantly lower (see Table 2).

A significant effect of gait speed was found on CoA_ext and CoA_flex (see Table 2). The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values CoA_ext was significantly higher, whereas the value of CoA_flex was significantly lower (see Table 2).

Figure 5 shows the rostro-caudal (from L3 to S2 spinal level) maps of lower limb muscle coactivation from slow walking to running with the averages TMCf_L3, TMCf_L4, TMCf_L5, TMCf_S1, TMCf_S2 (black lines), TO_e (black dots), and FWHM_L3, FWHM_L4, FWHM_L5, FWHM_S1, FWHM_S2 (the area underlying the curves, red and grey for run and walk respectively) of all 19 runners.

A significant effect of gait speed was found on CI_L3, CI_L4, CI_L5, CI_S1, CI_S2, on the Max_L3, Max_L4, Max_L5, Max_S1, Max_S2, on the FWHM_L3, FWHM_L4, FWHM_L5, FWHM_S1, FWHM_S2, CMC_Wt_L3, CMC_Wt_L4, CMC_Wt_L5, CMC_Wt_S1, CMC_Wt_S2 (see Tab 2). The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values of CI_L4, CI_L5, CI_S1, CI_S2, Max_S1, Max_S2, CMC_Wt_S1, CMC_Bt_L3, CMC_Bt_L4, CMC_Bt_S1 and CMC_Bt_S2 were significantly higher, whereas the value of CMC_Bt_L5 were significantly lower (see Table 2).

A significant effect of gait speed and level were found on CoA_L3, CoA_L4, CoA_L5, CoA_S1, CoA_S2 (see Table 2 and Table 3). The post-hoc analysis on speed revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values CoA_L3, CoA_L4 and CoA_L5 were significantly higher, whereas the value of CoA_S1 and CoA_S2 were significantly lower (see Table 2). The post-hoc analysis on spine levels revealed that at the transition between walking (6.8 Km/h) and running (7.3 Km/h) the values CoA_S1 were significantly higher than CoA_L3, CoA_L4, CoA_L5, as well as the values CoA_S2 were significantly higher than CoA_L3, CoA_L4, CoA_L5 (see Table 3).

3.2. Cross-Correlation

A significantly higher shape similarity between the global and the extensor coactivation map then global and flexor coactivation map was found (see Table 4).

A significant effect of shape similarity was found between the global and rostro-caudal coactivation maps (see Table 4). The post-hoc analysis revealed that the value of shape similarity between global and S1 coactivation map was significantly higher than global and L3, L5 and S2 coactivation maps, as well as the shape similarity between global and L4 coactivation map was significantly higher than global and L3, S2 coactivation maps (see Table 4).

3.3. Center of Mass Displacement and Spatiotemporal parameters

Figure 6 shows three-dimensional CoM maps in the vertical (CoM_y) and medio-lateral (CoM_z) directions. The maps were created using the CoM_y and CoM_z displacement average curves of 19 runners calculated on the gait cycle, from 0% to 100% (x-axis), for each speed of walking and running performed, from 0.8 to 9.3 km/h (y-axis) and with a variable amplitude (z-axis). A significant effect of the speed was found on CoM_y and on the CoM_z. The post-hoc analysis revealed that at the transition between walking (6.8 km/h) and running (7.3 km/h) the values of CoM_y displacement was significantly higher, whereas the value of CoM_z displacement was significantly lower (see Figure 6, and Table 2).

A significant effect of the speed was found on TO_e, stride length, stride frequency, foot lift. The post-hoc analysis revealed that the value of stride length, foot lift and stride frequency were significantly higher, whereas the values of TO_e, was significantly lower at the transition between walking (6.8 km/h) and running (7.3 km/h) (see Figure 7, and Table 2).

3.4. Correlations

Regardless of gait speed, CoM_y was positively correlated with higher CI_glob (r = 0.88, p < 0.001) and Max_glob (r = 0.89, p < 0.001), and negatively correlated with FWHM_glob (r = -0.83, p < 0.001) values, whereas CoM_z was positively correlated with CoA_glob values (r = 0.97, p < 0.001). Stride length values were negatively correlated with CI_glob, (r = - 0.98, p < 0.001) and Max_glob (r = -0.99, p < 0.001) values, and positively correlated with FWHM_glob (r = 0.80, p < 0.001). Foot lift values were positively correlated with Max_glob values (r = 0.72, p = 0.001), cadence values were negatively correlated with CoA_glob values (r = - 0.72, p < 0.001), whereas TO_e values were positively correlated with CoA_glob values (r = 0.79, p < 0.001).

4. Discussion

The objective of this study was to explore the behavior of the lower limb muscles' coactivation across several speeds, ranging from walking to running, as an expression of the CNS's global strategy for controlling the simultaneous activation of several lower limb muscles during locomotion.

We found that at higher speeds, the global coactivation index rises with increasing coactivation function values, and that coactivation occurs earlier and for a shorter period than at slower speeds. This is especially evident while transitioning from walking to running, as indicated by higher CI_glob and Max_glob values when walking and lower CoA_glob and FWHM_glob values while running (see Figure 2 and Figure 3). We also found that running resulted in increased vertical displacement, foot lift, stride frequency, and decreased lateral displacement and stride length, compared to walking (see Figure 7, Figure 6), which is consistent with earlier research on the biomechanics of running [18,19,20,21]. Interestingly, regardless of gait speed, global coactivation co-activation occurring earlier and in a confined portion of the gait cycle was positively correlated with higher vertical displacement, stride frequency, foot lift, and negatively correlated with lower lateral displacement and stride length.

From slow walking to running, we found a different set of curve shapes in lower limb muscle coactivation (see Figure 1). The time-varying multi-muscle coactivation function revealed a one-hump configuration during slow walking, a four-hump configuration during fast walking, and a three-hump configuration during running. The lonely hump we found during slow walking, as well as the first hump we discovered during running, paralleled the entire duration of the stance contact phase, whereas the first two humps of fast walking distinctly corresponded to the loading (in the range of 10-20% of the gait cycle) and push-off (in the range of 40-60% of the gait cycle) subphases, respectively.

Overall, these humps reflected an increase in whole-limb stiffness in response to ground impact for weight acceptance and pushing on the ground for propulsive purposes, which was consistent with previous findings that the spatiotemporal profile of global coactivation matches that of ground reaction force (GRF) [31,37]. Both gait variability and CoM oscillation are known to increase during slow walking, resulting in less stable walking compared to faster gait [9,63,64]. The longer duration, lower magnitude (see Figure 1), and higher within and between subject variability of lower limb global coactivation observed at slow walking, as expressed by CMC values (See Figure 2), suggest that the increase in whole-limb stiffness is exerted within a long-lasting and variable "safety strategy," primarily designed to maintain dynamic balance during body progression [63,64].

Conversely, the shorter duration, higher magnitude (see Figure 1), and lower within and between-subject variability of lower limb global coactivation (see Figure 2) observed in running suggest that whole-limb stiffness is exerted in a more synchronized and stable muscle activation. It also implies that when transitioning from walking to running, the increased whole-limb muscle coactivation during the foot contact phase is intended to absorb ground impact and generate propulsive forces in a single motor act, most likely within a unique motor strategy. As a result, the CI could be a useful index for reflecting and assessing the efficiency of the "leg-spring" stiffness mechanism, which is correlated with a more economical running [65,66].

The findings of the partial correlation analysis support the observed mechanisms of global coactivation. Regardless of gait speed, earlier and shorter values of global coactivation during the gait cycle reduce stride length while increasing vertical and decreasing lateral CoM displacements. As a result, rather than being the result of changes in gait speed, it is possible to hypothesise that the observed differences in spatiotemporal and kinematic variables between walking and running could be the result of global coactivation, which is a sensory-control integration process used by the CNS to deal with a more demanding and potentially unstable task like running [67,68,69]. This explains why similar results in terms of coactivation indexes and kinematic characteristics were found in subjects with abnormal gait stability control, such as cerebellar ataxia, who attempt to compensate for gait instability by using global coactivation [37,56,58].

When we analysed the coactivation of either flexor or extensor muscles separately, we found that the function curve of the extensor muscles matched that of the whole limb, whereas that of the flexor muscles clearly differed from it (see Figure 4). These findings suggest that the main contribution to the whole limb stiffening during walking and running is mainly given by the coactivation of the extensor muscles according to their role in weight acceptance and propulsive function [70]. Such an extensors activation clearly corresponds to the first hump of the global coactivation of both running and walking (both slow and fast) curves, as well as to the second hump of fast walking. Conversely, the coactivation curve of the flexor muscles increased during the early step air phase in line with the role the flexor muscles in limb lifting [70]. The coactivation of the flexor muscles, although small, was clearly present in running as second hump of the global coactivation curve and in fast walking as third hump (see Figure 4). This last result is of particularly interest because it suggests that the flexor muscles is a key factor in transitioning from fast walking to running. It also fits well with the observation of a propagation delay in muscle coactivation from L3 to S2 (see Figure 5). These findings are in line with previous findings on walking [46,71,72] and reinforce the hypothesis that the coactivation of the flexor muscles (e.g., rectus femoris, iliopsoas), whose motoneurons are located more rostrally within the spinal cord, may play a crucial role in the switch from walking to running. Interestingly, although there was a clear progressive increase in the amount of coactivation of either global extensors or flexors from slow to fast walking, we found an increase in flexor coactivation just before the running speed threshold in the late fast walking trials, followed by a mild reduction in coactivation in the early running trials (see Figure 4 and Table 2). One of the key features in the passage from fast walking to running is the optimal exploitation of the mass-spring mechanism to generate kinetic energy in running [15,17]. We might infer that such an advantage, which occurs during ground contact, is related to the global coactivation, and reduces the need to coactivate the flexor muscles during the step air phase in early running compared to late fast walking. It is important to note that the coactivation of the flexor muscles corresponded to a small peak (second hump) in the “hollow” of the global coactivation curve, possibly suggesting that the flexor muscles need to be coactivated when the limb stiffness is reduced to facilitate the forward progression.

This study has several limitations. First, due to the experimental setup, antero-posterior CoM displacements could not be assessed, and thus kinetic energy parameters could not be calculated. We also did not account for oxygenation parameters, so our data cannot provide a complete characterization of global coactivation on running economy. However, because greater neuromuscular activation, vertical stiffness, and the ability to rapidly produce force throughout the lower limb during ground contact have been shown to correlate with more economical running [66,73,74], it is expected that CI values may correlate with the energetic running profile in future studies evaluating running performances. Another limitation is related to the superimposition of gait speeds through the use of treadmill, which might have affected some gait parameters [75]. Therefore, further studies examining global coactivation indices during overground running are needed.