Effects of spinal transection and locomotor speed on muscle synergies of the cat hindlimb

Abstract

Abstract

It has been suggested that during locomotion, the nervous system controls movement by activating groups of muscles, or muscle synergies. Analysis of muscle synergies can reveal the organization of spinal locomotor networks and how it depends on the state of the nervous system, such as before and after spinal cord injury, and on different locomotor conditions, including a change in speed. The goal of this study was to investigate the effects of spinal transection and locomotor speed on hindlimb muscle synergies and their time‐dependent activity patterns in adult cats. EMG activities of 15 hindlimb muscles were recorded in nine adult cats of either sex during tied‐belt treadmill locomotion at speeds of 0.4, 0.7 and 1.0 m/s before and after recovery from a low thoracic spinal transection. We determined EMG burst groups using cluster analysis of EMG burst onset and offset times and muscle synergies using non‐negative matrix factorization (NNMF). We found five major EMG burst groups and five muscle synergies in each of six experimental conditions (2 states × 3 speeds). In each case, the synergies accounted for at least 90% of muscle EMG variance. Both spinal transection and locomotion speed modified subgroups of EMG burst groups and the composition and activation patterns of selected synergies. However, these changes did not modify the general organization of muscle synergies. Based on the obtained results, we propose an organization for a pattern formation network of a two‐level central pattern generator that can be tested in neuromechanical simulations of spinal circuits controlling cat locomotion.

Key points

• Analysis of muscle synergies during locomotion can reveal the organization of spinal locomotor networks.

• We recorded EMG activity of 15 hindlimb muscles in cats locomoting on a treadmill at speeds 0.4, 0.7 and 1.0 m/s before and after recovery from spinal cord transection at low thoracic level.

• We found five muscle synergies in all six experimental conditions (2 spinal states x 3 speeds) that include two flexor synergies operating in the swing phase and three extensor synergies operating in the stance phase.

• Major features of found synergies (the number, muscle composition and activation patterns) were not substantially affected by spinal transection and locomotion speed, suggesting that spinal control mechanism operates muscle synergies.

• Based on the obtained results, we proposed an organization of a pattern formation network of a two‐level central pattern generator controlling locomotor activity of hindlimb muscles.

Abstract figure legend Locomotor electromyographic (EMG) activity of hindlimb muscles of intact and spinal cats reveals muscle synergies and suggests organization of spinal locomotor networks.

Introduction

A common morphological feature of limb design in mammals is muscle redundancy (or abundance), defined as a seemingly excessive number of muscles with respect to the number of kinematic degrees of freedom of the limb. For example, the hindlimbs of rodents, cats and dogs have over 30 muscles serving seven major degrees of freedom, with three at the hip, one or two at the knee and two or three at the ankle (Burkholder & Nichols, 2004; Charles et al., 2016; Ramalingasetty et al., 2021; Stark et al., 2021). It has been suggested that the nervous system simplifies its control of multiple muscles by combining them in muscle groups, termed motor modules or synergies, and activating each of these groups with a specific time‐dependent activation pattern (Bernstein, 1967; Bernstein, 2021; Bizzi et al., 2008; Cheung & Seki, 2021; d'Avella et al., 2006; Giszter, 2015; Ivanenko et al., 2005; Lacquaniti et al., 2012). Besides synergies revealed at the level of individual muscles, there is ample evidence for the existence of motoneuronal synergies within and across muscles that receive shared or unique synaptic input (De Luca & Erim, 1994; Del Vecchio et al., 2023; Gibbs et al., 1995; Levine et al., 2023; Madarshahian et al., 2021).

Analyses of muscle and motoneuronal synergies during locomotion can potentially reveal the organization of spinal premotor interneuronal networks controlling motoneuronal locomotor activity, called central pattern generators (CPG). Currently, there is little consensus on the organization of mammalian locomotor CPGs with several possible schemes proposed. These include unit burst generators (Grillner, 1981; Grillner & Kozlov, 2021), a two‐level CPG comprising a rhythm generator (RG) and a pattern formation (PF) network (Burke et al., 2001; McCrea & Rybak, 2008), a nested CPG organization (Berkowitz, 2019), multistable half‐centre oscillators (Bondy et al., 2016; Parker et al., 2021), and networks with recurrent excitatory and inhibitory connectivity generating rotational dynamics (Linden et al., 2022). Muscle synergies identified in locomotion of rodents (DiGiovanna et al., 2016; Santuz et al., 2019), cats (Desrochers et al., 2019; Harnie et al., 2021; Klishko et al., 2021; Krouchev et al., 2006; Markin et al., 2012) and humans (Ivanenko et al., 2005; Monaco et al., 2010; Saito et al., 2018) indicate the existence of a relatively small number of muscle groups activated together by common inputs, consistent with several versions of CPG organization.

Several factors, such as locomotion speed and the state of the nervous system (e.g. following spinal cord injury), have been reported to affect the number and composition of muscle locomotor synergies. In healthy children (Rozumalski et al., 2017) and younger and older adults (Dewolf et al., 2019; Ivanenko et al., 2003; Monaco et al., 2010; Saito et al., 2018), walking speeds do not affect the number of muscle synergies and have low to moderate effects on the composition and activation patterns of synergies within an age group.

Spinal cord injury appears to affect the number and composition as well as the activation patterns of muscle synergies. For instance, children and adults with incomplete spinal cord injury on average have fewer synergies (2–4) and different synergy composition and activation patterns compared to four to five muscle synergies of healthy age‐matched humans (Fox et al., 2013; Hayes et al., 2014; Sun et al., 2022). Another study reported five muscle synergies in individuals with incomplete spinal cord injury that were similar in composition and patterns to the synergies in a healthy control group (Ivanenko et al., 2003). All the above studies reported greater variability of synergy activation patterns after spinal cord injury. The discrepancies in the number and composition of muscle synergies and their activation patterns in studies of patients with spinal cord injury could be explained by differences in the severity and location of the spinal injury, the number of studied muscles, the type of walking assistance and rehabilitation interventions. These differences make it difficult to reveal consistent muscle synergies and a potential organization of spinal locomotor CPGs that are likely to operate not only in quadrupedal mammals (Grillner & Kozlov, 2021; Kiehn, 2016; McCrea & Rybak, 2008) but also in humans (Duysens & Van de Crommert, 1998; Minassian et al., 2023; Shapkova, 2004).

Two recent studies in cats with low‐thoracic spinal transection (i.e. spinal cats) have revealed up to seven muscle groups that are activated together in the walking cycle and are generally similar to the groups during intact locomotion (Desrochers et al., 2019; Higgin et al., 2020). These muscle groups were identified based on the burst onset and offset times of electromyographic (EMG) activity of several hindlimb muscles using cluster analysis and provided initial information about the potential organization of spinal locomotor networks. Specifically, in both spinal and intact cats, there are muscle activity burst groups comprising either flexor or extensor muscles and some of these groups contain muscles operating at different joints (Desrochers et al., 2019; Harnie et al., 2021; Higgin et al., 2020; Klishko et al., 2021; Krouchev et al., 2006; Markin et al., 2012). These results suggest that all muscles within individual flexor and extensor burst groups receive a common input from a flexor and an extensor CPG half‐centre, respectively. However, the above cluster analysis cannot determine the activity contribution of individual muscles within a group or the pattern of the activation input to each group.

Therefore, the goal of this study was to determine hindlimb muscle synergies and their activation patterns in cats before and after a low‐thoracic spinal transection during locomotion at different speeds on a tied‐belt treadmill. We tested the hypothesis that a spinal mechanism controls the number, composition and activity patterns of muscle synergies as a function of locomotor speed.

Methods

Ethical approval

All procedures were approved by the Animal Care Committee of the Université de Sherbrooke (Protocol 442‐18) in accordance with policies and directives of the Canadian Council on Animal Care. We obtained the current data set from nine adult purpose‐bred cats (>1 year of age at the time of experimentation), four females and five males, weighing between 3.4 and 4.8 kg, purchased from Marshall BioResources (North Rose, NY, USA). Before and after experiments, cats were housed and fed (weight‐dependent metabolic diet and water ad libitum) in a dedicated room within the animal care facility of the Faculty of Medicine and Health Sciences at the Université de Sherbrooke. We followed the ARRIVE guidelines 2.0 for animal studies (Percie du Sert et al., 2020). The investigators understand the ethical principles under which the journal operates, and our work complies with its animal ethics checklist. In order to maximize the scientific output of each animal, they were used in other studies to investigate different scientific questions, some of which have been published (Audet et al., 2022; Harnie et al., 2021; Harnie et al., 2024; Mari et al., 2023).

Surgical procedures

The implantation and spinal transection surgeries were performed under aseptic conditions with sterilized equipment in an operating room. Prior to surgery, cats were sedated with an intramuscular (i.m.) injection of butorphanol (0.4 mg/kg), acepromazine (0.1 mg/kg), and glycopyrrolate (0.01 mg/kg). We then injected a mixture (0.05 ml/kg, i.m.) of diazepam (0.25 mg/kg) and ketamine (2.0 mg/kg) in a 1:1 ratio 5 min later for induction. We shaved the animal's fur (back, stomach, fore‐ and hindlimbs) and cleaned the skin with chlorhexidine soap. Cats were anaesthetized with isoflurane (1.5–3%) and O₂ delivered with a mask and then with a flexible endotracheal tube. The depth of anaesthesia was confirmed by applying pressure to a paw (to detect limb withdrawal) and by assessing the size and reactivity of pupils. Isoflurane concentration was adjusted throughout the surgery by monitoring cardiac and respiratory rates. Body temperature was maintained constant (37 ± 0.5°C) using a water‐filled heating pad placed under the animal, an infrared lamp placed ∼50 cm over it and a continuous infusion of lactated Ringer's solution (3 ml/kg/h) through a catheter placed in a cephalic vein. At the end of surgery, we injected subcutaneously an antibiotic (cefovecin, 8 mg/kg) and a fast‐acting analgesic (buprenorphine, 0.01 mg/kg). We also taped a fentanyl (25 µg/h) patch to the back of the animal 2–3 cm rostral to the base of the tail for prolonged analgesia, which we removed 4–5 days later. After surgery, the cats were placed in an incubator and closely monitored until they regained consciousness. We administered another dose of buprenorphine ∼7 h after surgery.

To record EMG, we directed pairs of Teflon‐insulated multistrain fine wires (AS633; Cooner Wire Co., Chatsworth, CA, USA) subcutaneously from two head‐mounted 34‐pin connectors (Omnetics Connector Corp., Minneapolis, MN, USA). Two wires, stripped of 1–2 mm of insulation, were sewn into the belly of selected hindlimb muscles for bipolar recordings. The head‐mounted connectors were fixed to the skull using dental acrylic and four to six screws. We verified electrode placement during surgery by stimulating each muscle through the appropriate head connector channel to assess the biomechanically desired muscle contraction. During experiments, EMG signals were pre‐amplified (×10, custom‐made system), bandpass filtered (30–1000 Hz) and amplified (100–5000×) using a 16‐channel amplifier (model 3500; AM Systems, Sequim, WA, USA). As we implanted more than 16 muscles per cat, we obtained data in each locomotor condition twice, one for each connector, as our current data acquisition system is limited to 16 channels. EMG data were digitized (5000 Hz) with a National Instruments (Austin, TX, USA) card (NI 6032E), acquired with custom‐made acquisition software and stored on computer. We implanted the following hindlimb muscles bilaterally: flexor digitorum longus (FDL, digits and ankle plantarflexor), peroneus longus (PLO, ankle abductor and dorsiflexor), soleus (SO, ankle plantarflexor), lateral gastrocnemius (LG, ankle plantarflexor and knee flexor), medial gastrocnemius (MG, ankle plantarflexor and knee flexor), plantaris (PL, digits and ankle plantarflexor and knee flexor), vastus lateralis (VL, knee extensor), biceps femoris anterior (BFA, hip extensor), gluteus (GLU, hip abductor and extensor), caudofemoralis (CF, hip abductor and extensor), biceps femoris posterior (BFP, hip extensor and knee flexor), semitendinosus (ST, hip extensor and knee flexor), tibialis anterior (TA, ankle dorsiflexor), sartorius anterior (SRTa, hip flexor and knee extensor) and iliopsoas (IP, hip flexor). We used EMG data from both or either left or right hindlimbs, depending on signal quality.

One to two weeks after electrode implantation, we collected data in the intact state for 4–6 weeks. A complete spinal transection was then made at low thoracic levels. Before surgery, we sedated the cat with an intramuscular injection of a cocktail containing butorphanol (0.4 mg/kg), acepromazine (0.1 mg/kg) and glycopyrrolate (0.01 mg/kg) and inducted with another intramuscular injection (0.05 ml/kg) of ketamine (2.0 mg/kg) and diazepam (0.25 mg/kg) in a 1:1 ratio. We shaved the fur overlying the back, stomach and hindlimbs and cleaned the skin with chlorhexidine soap. The cat was then anaesthetized with isoflurane (1.5–3%) and O₂ using a mask for a minimum of 5 min and then intubated with a flexible endotracheal tube. Isoflurane concentration was confirmed and adjusted throughout the surgery by monitoring cardiac and respiratory rates, by applying pressure to the paw to detect limb withdrawal and by assessing muscle tone. Once the animal was deeply anaesthetized, the skin was incised over the 12th and 13th (T12–T13) thoracic vertebrae, and after setting aside muscle and connective tissue, a small laminectomy of the dorsal bone was made. After exposing the spinal cord, we applied xylocaine (lidocaine hydrochloride, 2%) topically and made two to three injections within the spinal cord. We then completely transected the spinal cord with surgical scissors. We then cleaned the ∼0.5 cm gap between the two cut ends of the spinal cord and stopped any residual bleeding. We verified that no spinal cord tissue remained connecting rostral and caudal ends, which we later confirmed histologically. A haemostatic material (Spongostan) was inserted within the gap, and muscles and skin were sewn back to close the opening in anatomical layers. At the end of surgery, we injected an antibiotic (Cefovecin, 0.1 ml/kg) subcutaneously and taped a transdermal fentanyl patch (25 µg/h) to the back of the animal 2–3 cm rostral to the base of the tail to provide prolonged analgesia, which was removed 4–5 days later. We also injected buprenorphine (0.01 mg/kg), a fast‐acting analgesic, subcutaneously at the end of the surgery and a second dose ∼7 h later. After surgery, we placed the cat in an incubator until it regained consciousness. After spinal transection, we manually expressed the cat's bladder and large intestine 2–3 times daily. Cats were then monitored daily by experienced personnel. The hindlimbs were cleaned as needed to prevent infection. Following surgery, once the animal was fully conscious (capable of holding its head upright, pupils no longer dilated, calm behaviour), we mixed a can of moist food with water and left it for the animal's consumption and rehydration. Dry food was then offered the next day. In the 8 h after waking up, the animal was closely monitored, repositioned, and stimulated to eat and drink. To monitor the state of hydration of the animal, we performed skin fold tests, tested for capillary refill time and verified the urine (e.g. colour, consistency) when manually voiding the bladder. If the animal was not sufficiently hydrated, we performed and maintained fluid therapy (200 ml/day of NaCl 0.9%, s.c.) until hydration returned to normal. The animals were closely monitored, and their general state (respiratory and cardiac rates, temperature and overall behaviour) was evaluated by qualified personnel three times per day. At 5–7 days post‐transection, animals were fully independent for feeding, drinking and maintaining their body temperature within normal values (37 ± 0.5°C), as well as moving around using their forelimbs. At the end of experiments, cats were anaesthetized with isoflurane (1.5–3.0%) and O₂ before receiving a lethal dose of pentobarbital (120 mg/kg) through the left or right cephalic vein. Cardiac arrest was confirmed using a stethoscope to determine the death of the animal.

Experimental design and data collection

Data collection in spinal cats began between the 8th and 15th weeks after spinal transection, based on the recovery of a robust hindlimb locomotion in the forward direction with full hindquarter weight support without or with perineal stimulation. For perineal stimulation, the experimenter manually pinched or rubbed the skin under the tail with the index finger and thumb. During data collection in spinal cats, the experimenter held the tail of the animal to provide mediolateral balance assistance but did not provide weight support. We did not train spinal cats to recover hindlimb locomotion as discussed previously (Harnie et al., 2019). Cats performed hindlimb‐only locomotion with the forelimbs on a stationary platform on a split‐belt treadmill consisting of two independently controlled running surfaces 120 cm long and 30 cm wide (Bertec, Columbus, OH, USA). A Plexiglas separator (120 cm long, 3 cm high, and 0.5 cm wide) was placed between the left and right belts to prevent the limbs from impeding each other. Cats stepped by the left and right hindlimb on the left and right belts that moved at equal speeds of 0.4, 0.7 and 1.0 m/s (tied‐belt treadmill locomotion). Tied‐belt locomotion on a split‐belt treadmill is slightly different from a typical treadmill locomotion, that is, on the split‐belt treadmill animal left and right limbs are constrained to step on the left and right belt, respectively. At each speed, the objective was to obtain a minimum of 10–15 consecutive locomotion strides (cycles).

We collected kinematic data by capturing videos of the left and right sides using two cameras (Basler AcA640‐100 g, Basler AG, Germany) at 60 frames/s with a spatial resolution of 640 × 480 pixels. A custom‐made program (LabVIEW, National Instruments) acquired the images and synchronized acquisition with EMG data. We identified stance phase onset and offset times by visual inspection of video recordings. We also digitised at least 10–15 video recordings for each cat and experimental condition. Using these digitized recordings, we measured stride length for the right hindlimb as the distance between stance onset and offset added to the distance travelled by the treadmill during the swing phase, which was calculated by multiplying swing duration by treadmill speed (Courtine & Schieppati, 2004; Dambreville et al., 2015; Thibaudier & Frigon, 2014).

Analysis of muscle activity, burst groups and synergies

We described the analysis of EMG activity and determination of EMG burst groups and muscle synergies previously (Harnie et al., 2021; Klishko et al., 2021; Markin et al., 2012) and therefore provide a shorter description here. We rectified raw EMG signals and determined onset and offset times of each burst using an EMG threshold (typically the mean EMG interburst baseline plus 2 standard deviations (SD)). Occasionally (in less than 10% of analysed cycles), when the EMG baseline was noisy (approaching ∼20% of the mean EMG burst magnitude), we increased the threshold up to 3 SD. In three experimental conditions of two cats (HA, intact state, speeds 0.7 and 1.0 m/s and SA, intact state, speed 1.0 m/s; Table 1) the quality of signal was low, and we were unable to reliably identify muscle bursts. We computed the mean amplitude of each EMG burst and normalized it to the maximum mean burst magnitude determined for each EMG channel and cat across all experimental conditions.

Table 1.

Animal characteristics and number of investigated locomotion cycles in all experimental conditions

			Number of locomotion cycles
			Intact			Spinal
Cat	Sex	Mass (kg)	0.4 m/s	0.7 m/s	1.0 m/s	0.4 m/s	0.7 m/s	1.0 m/s	Total
AM	F	3.6	20	42	29	60	51	33	235
C3	M	4.7	45	59	46	29	62	28	269
CR	F	4.1	28	31	27	12	15	16	129
DA	M	4.4	39	53	58	44	39	38	271
DO	F	3.6	50	71	76	60	69	75	401
HA	M	4.0	51	16	16	51	34	40	176
SA	F	3.6	46	57	12	132	73	76	384
SN	M	4.8	64	79	72	59	60	40	374
TO	M	3.4	31	36	30	27	65	29	218
Mean ± SD or total		4.02 ± 0.51	374	444	366	474	468	375	2501

Note: The number of cycles in the table was used for kinematic and EMG analyses. In three experimental conditions of two cats (cycle numbers in bold), the quality of EMG recordings was low and these cycles were used only for kinematic analysis.

We used two methods to determine muscle synergies from EMG recordings. The first method was proposed in Krouchev et al. (2006) and subsequently used in other studies (Desrochers et al., 2019; Higgin et al., 2020; Klishko et al., 2021; Markin et al., 2012; Yakovenko & Drew, 2015). This method defines muscle synergies as muscle groups that are activated in the same phase of the locomotor stride (cycle), that is, whose EMG burst onset and offset times are synchronised. We first determined onset and offset times of each muscle burst with respect to the start of the corresponding locomotor cycle, that is, the swing phase onset. We then expressed each EMG burst onset and offset time as the phase of the locomotion cycle by normalizing them to the corresponding cycle duration; the cycle duration was defined as the period between the onset times of two subsequent swing phases. EMG bursts of all muscles, cats and cycles in each experimental condition were presented as a scatter cloud of points where each EMG burst offset phase is plotted as a function of the corresponding EMG burst onset phase. We performed a cluster analysis of EMG burst clatter points using the minimum spanning tree algorithm in MATLAB (MathWorks, Natick, MA, USA) for each combination of spinal cord state (intact and spinal‐transected) and locomotion speed (0.4, 0.7 and 1.0 m/s). In this clustering algorithm, a set of EMG bursts for each experimental condition was represented by an edge‐weighted graph. The graph vertices represented the centres of burst distribution of each muscle computed as the means of muscle burst onsets and offsets. The graph edges were obtained as the weighted distances between pairs of muscle burst centres expressed in standard deviations. The algorithm determined the minimal spanning tree of an edge‐weighted graph in which the total weights of all edges did not exceed total weights of any other spanning tree. Subgraphs of the minimal spanning tree, representing clusters of EMG bursts activated together, were obtained by removing edges with largest weights connecting the subgraphs (Markin et al., 2012; Vathy‐Fogarassy et al., 2005). Muscles whose EMG bursts were in the same cluster, that is, activated together, were considered synergists. Although muscle synergies defined this way have the straightforward interpretation, their magnitude and pattern of activity are not obtained.

Another frequently used method of synergy analysis, a non‐negative matrix factorization (NNMF), defines synergies as groups of muscles where each muscle makes a specific relative contribution to the group activity magnitude and all muscles in the group (synergy) have the same time‐dependent activation pattern (Cheung et al., 2005; Ivanenko et al., 2005; Tresch & Jarc, 2009). To determine muscle synergies using this approach, we first subtracted EMG values below the burst threshold from the raw rectified EMG within each trial and smoothed the signal using a Butterworth fourth order zero‐lag digital filter with a cutoff frequency of 10 Hz. We used a NNMF algorithm (MATLAB function nnmf with parameter replicates = 20) to compute muscle weights (the relative contribution of each muscle to individual synergies, matrix W) and time‐dependent activation of each synergy (matrix C) from the following equation: D = WC + error. Here D is a (m $\times$ T) matrix of EMG envelope values for each muscle and percentage of locomotion cycle; m = 15 is the number of muscles; T = 101 is the number of normalized time samples in the cycle (0, 1, …, 100); W is a (m $\times$ n) matrix of muscle weights indicating each muscle contribution to each synergy; n is the number of synergies; C is a (n $\times$ T) matrix of time‐dependent activation coefficients for each synergy; and error is a (m $\times$ T) matrix of deviations of matrix D from matrix D* = WC. We generated 50 matrices D by randomly selecting cycles of EMG envelopes for each muscle across all cats with available cycles within each experimental condition (Table 1) to maximize representation of different cats in computed synergies (Klishko et al., 2021). We computed matrices W and C for each experimental condition using a different number of synergies (from n = 1 to n = 14). We did not perform computations for n = 15 because this number of synergies corresponded to the number of muscles and would give a trivial solution in which D = D*, that is, EMG patterns reconstructed from the obtained synergies would be the same as the recorded patterns. We selected for further analysis the smallest number of synergies that accounts for at least 90% of the mean EMG variance computed across all 50 matrices D within each condition.

To evaluate the similarity of muscle synergies in different experimental conditions, we also computed matrices W _COM and C _COM for different combinations of spinal states and speeds and evaluated the variance accounted for by their products D^* _COM = W _COM C _COM in the corresponding experimental EMG patterns D _COM (Harnie et al., 2021; Klishko et al., 2021). Here the subscript COM denotes a specific combination of spinal states and speeds. For example, we evaluated the effect of spinal cord state on muscle synergies at given locomotion speeds by computing matrices ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4}^{\ast }={\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.4}{\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$, ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.7}^{\ast }={\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.7}{\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$, and ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_1.0}^{\ast }={\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_1.0}{\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$ and calculated the variance accounted for by these matrices ${\mathbf{D}}_{\mathrm{COM}}^{\ast }$, containing reconstructed EMG patterns from muscle synergies, with the corresponding experimental EMG patterns (matrices ${\mathbf{D}}_{\mathrm{In}\_0.4}$, ${\mathbf{D}}_{\mathrm{Sp}\_0.4}$, etc.). The subscript In $|$ Sp in the above example indicates a combination of the intact and spinal states; the subscripts 0.4, 0.7 and 1.0 correspond to locomotion speeds. In other words, matrices ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$, ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$ and ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$ are composed of all intact and spinal EMG patterns for each speed (Fig. 1 top, middle and bottom panels, respectively). The dimensions of these and corresponding C _COM matrices are (m $\times$ 2T). The dimensions of the corresponding matrices W _COM are (m $\times$ n). Similarly, we evaluated the effect of locomotion speed on muscle synergies in the intact and spinal states by computing matrices ${\mathbf{D}}_{\mathrm{In}\_0.4|0.7|1.0}^{\ast }={\mathbf{W}}_{\mathrm{In}\_0.4|0.7|1.0}{\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$ and ${\mathbf{D}}_{\mathrm{Sp}\_0.4|0.7|1.0}^{\ast }={\mathbf{W}}_{\mathrm{Sp}\_0.4|0.7|1.0}{\mathbf{C}}_{\mathrm{Sp}\_0.4|0.7|1.0}$, where the subscript $0.4|0.7|1.0$ indicates a combination of all locomotion speeds together. The corresponding matrices of experimental EMG patterns (${\mathbf{D}}_{\mathrm{In}\_0.4|0.7|1.0}$ and ${\mathbf{D}}_{\mathrm{Sp}\_0.4|0.7|1.0}$) are composed of all EMG patterns on the left and right panels of Fig. 1, respectively. The dimensions of these and the corresponding C _COM matrices are (n $\times$ 3T); the corresponding matrices W _COM have dimensions (m $\times$ n). Finally, we computed matrices ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}^{\ast }={\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}{\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}$ that included all six experimental conditions. Matrices ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}^{\ast }$ and ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}$ have dimensions (n $\times$ 6T). If matrices W _COM and C _COM for different combinations of spinal states and speeds reproduced the corresponding experimental EMG patterns D _COM with the same accuracy (variance accounted for), we would conclude that matrices W _COM and C _COM match matrices W and C for each of the six experimental conditions (2 spinal states × 3 speeds) and thus the spinal state and locomotion speed do not affect the number, composition and patterns of muscle synergies.

Figure 1. Low‐pass filtered patterns of EMG activity of hindlimb muscles during intact (left panels) and spinal (right panels) tied‐belt treadmill locomotion at speeds 0.4 m/s (top panels), 0.7 m/s (middle panels) and 1.0 m/s (bottom panels).

Grey lines show EMG patterns of 2457 analysed cycles of 9 cats (see Table 1); thick black lines are the mean EMG patterns across all cycles and cats within a muscle and experimental condition. The EMG magnitude of each muscle is normalized to the peak low‐pass filtered EMG across all experimental conditions within the cat. The vertical axis in each plot designates the range of the normalized EMG magnitude from 0 to 1. The vertical line within each EMG panel separates the swing and stance phases. Extensors are muscles with a stance phase related activity and primary function of supporting body against gravity and propelling it forward: FDL, flexor digitorum longus (digits and ankle plantarflexor); PLO, peroneus longus (ankle abductor and a dorsiflexor); SO, soleus (ankle plantarflexor); LG, lateral gastrocnemius (ankle plantarflexor and knee flexor); MG, medial gastrocnemius (ankle plantarflexor and knee flexor); PL, plantaris (digits and ankle plantarflexor and knee flexor); VL, vastus lateralis (knee extensor); BFA, biceps femoris (hip extensor); GLU, gluteus (hip abductor and extensor); CF, caudofemoralis (hip abductor and extensor). Flexors are muscles with a swing phase related activity: TA, tibialis anterior (ankle dorsiflexor); SRTa, sartorius anterior (hip flexor and knee extensor); IP, iliopsoas (hip flexor). Two muscles (BFP, biceps femoris posterior, hip extensor and knee flexor) and ST (semitendinosus, hip extensor and knee flexor) are bifunctional muscles with swing and stance related activity.

Statistics

We evaluated the significance of the effects of independent factors of state (intact, spinal), locomotion speed (0.4, 0.7, 1.0 m/s), and/or muscle (m = 15) on dependent variables, such as temporal cycle characteristics, stride length, normalized EMG burst magnitude, and onset and offset times. In this statistical analysis, we considered animals as a random factor. We evaluated significance of main effects of the above factors and their interactions using a linear mixed‐effects model (MIXED function; IBM SPSS Statistics 29, IBM Corp., Armonk, NY, USA). We performed post hoc pairwise comparisons with Bonferroni adjustments.

We used a similar linear mixed‐effects model to test for the significance of the effects of state and locomotion speed on muscle weights (matrices W). A separate linear mixed‐effects model analysis was performed to evaluate the significance of the difference in synergy activation coefficients (matrices C) at each 10%‐time bin of the locomotor cycle. In addition, we computed the coefficient of determination R ² between pairs of mean activation patterns corresponding to each experimental condition (e.g. ${\mathbf{C}}_{\mathrm{In}\_0.4}$, ${\mathbf{C}}_{\mathrm{Sp}\_1.0}$, etc.) and the corresponding components of matrices C _COM (e.g. ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$, etc.). For all statistical tests, the significance was set at 0.05.

Results

General characteristics of EMG activity patterns and bursts

Figure 1 demonstrates experimentally obtained EMG linear envelopes of 15 hindlimb muscles in analysed locomotor cycles in the intact and spinal states at treadmill speeds of 0.4, 0.7 and 1.0 m/s. Hindlimb muscles with extensor (stance phase)‐related activity (top 10 muscles in each panel) initiated their activity at 10–15% and 20–25% of the cycle time prior to stance onset in intact and spinal state, respectively. The extensor activity bursts ended in the last 10–20% of the stance phase. Hindlimb muscles with flexor (swing phase)‐related activity (bottom three muscles) initiated their activity 10–20% prior to swing onset and the activity typically lasted until the end of swing with little or no‐coactivation with extensor muscles. Bifunctional BFP and ST muscles had flexor and extensor bursts at the stance–swing and swing–stance transitions, respectively, during intact locomotion at all speeds. EMG patterns of these muscles changed in the spinal state as the extensor burst became much smaller or disappeared, while the flexor burst increased in duration, especially in the ST. Two muscles with stance‐related activity, PLO and GLU, also had a flexor burst during swing in the intact state, which became much smaller or disappeared in the spinal state. PLO is an ankle abductor and dorsiflexor (Young et al., 1993) and GLU is a hip abductor and extensor (Crouch, 1969).

The normalized extensor burst magnitude across all muscles was generally greater in intact locomotion compared to spinal locomotion at 0.4 m/s (mean ± SD: 0.54 ± 0.21 vs. 0.50 ± 0.25), 0.7 m/s (0.60 ± 0.21 vs. 0.52 ± 0.27) and 1.0 m/s (0.71 ± 0.22 vs. 0.55 ± 0.25) (Fig. 2A ). The extensor burst magnitude also increased with locomotor speed. The effects of the state, speed and muscle independent factors were significant (F _1,5475 = 268.6, P < 0.001, F _2,5475 = 133.3, P < 0.001 and F _112,5475 = 148.3, P < 0.001, respectively). The interaction effect was also significant (F _60,5475 = 24.8, P < 0.001). We found no significant differences between the intact and spinal states for the extensor bursts of FDL (P = 0.891, P = 0.809), LG (P = 0.783, P = 0.986), and CF (P = 0.313, P = 0.455) at speeds 0.4 and 0.7 m/s (Fig. 2A ). In only one case did we observe an extensor muscle with a larger burst in the spinal state, the VL at a speed 0.4 m/s. The extensor bursts of PLO had a greater magnitude in the spinal state compared to the intact state at all speeds (P < 0.001). The SRTa did not have an extensor burst in the intact state at a speed of 0.4 m/s, the SRTa extensor burst magnitude was the same in the intact and spinal state at a speed of 0.7 m/s (P = 0.477), and its magnitude was larger after spinal transection compared to the intact state at 1.0 m/s (P < 0.001).

Figure 2. Bar plots and boxplots of normalized EMG burst magnitude of hindlimb muscles in intact and spinal states during locomotion at speeds 0.4, 0.7 and 1.0 m/s.

Bar plots with individual data points are shown for experimental conditions and muscles with n < 30. Boxplots are shown when n ≥ 30. The boxes in boxplots show the range of values between 25th and 75th percentiles with the median inside the boxes, and whiskers indicating minimal and maximal values. For muscle abbreviations see the legend for Fig. 1. A, normalized EMG extensor (stance related) burst magnitudes. Asterisks indicate statistical significance between intact and spinal states. For extensor bursts main effect of state was significant (P < 0.001, F _1,5475 = 268.6, linear mixed‐effects model). Number of data points for extensor bursts in the intact and spinal state are as follows. Speed 0.4 m/s: FDL, n = 16 and n = 12, respectively; PLO, n = 27 and n = 41; SO n = 139 and n = 161; LG, n = 66 and n = 107; MG, n = 58 and n = 163; PL, n = 71 and n = 71; VL, n = 73 and n = 146; BFA, n = 93 and n = 116; GLU, n = 25 and n = 37; CF, n = 28 and n = 75; BFP, n = 16 and n = 15; ST, n = 24 and n = 0; SRTa, n = 0 and n = 12. Speed 0.7 m/s: FDL, n = 13 and n = 25; PLO, n = 55 and n = 54; SO, n = 197 and n = 187; LG, n = 126 and n = 98; MG, n = 141 and n = 157; PL, n = 93 and n = 53; VL, n = 116 and n = 116; BFA, n = 164 and n = 116; GLU, n = 69 and n = 44; CF = 36 and n = 29; BFP, n = 67 and n = 36; ST, n = 66 and n = 24; SRTa, n = 12 and n = 15. Speed 1.0 m/s: FDL, n = 21 and n = 14; PLO, n = 54 and n = 46; SO, n = 135 and n = 139; LG, n = 98 and n = 63; MG, n = 114 and n = 148; PL, n = 91 and n = 43; VL, n = 124 and n = 75; BFA, n = 137 and n = 96; GLU, n = 56 and n = 15; CF, n = 46 and n = 24; BFP, n = 84 and n = 40; ST, n = 67 and n = 57; SRTa, n = 24 and n = 43. B, normalized EMG flexor (swing related) burst magnitudes. Asterisks indicate statistical significance between intact and spinal states. For flexor bursts the main effect of state was not significant (F _1,2035 = 2.427, P = 0.119), whereas the interaction effect of state‐speed‐muscle was significant (P < 0.001). Number of data points for flexor bursts in the intact and spinal state are as follows. Speed 0.4 m/s: TA, n = 26 and n = 20, respectively; PLO, n = 13 and n = 7; SRTa, n = 56 and n = 52; BFP, n = 77 and n = 50; ST, n = 81 and n = 15; GLU, n = 8 and n = 0; IP, n = 73 and n = 71. Speed 0.7 m/s: TA, n = 64 and n = 16; PLO = 32 and n = 6; SRTa, n = 65 and n = 38; BFP, n = 96 and n = 65; ST, n = 159 and n = 85; GLU, n = 12 and n = 0; IP, n = 61 and n = 30. Speed 1.0 m/s: TA, n = 77 and n = 20; PLO, n = 43 and n = 16; SRTa, n = 98 and n = 29; BFP, n = 105 and n = 95; ST, n = 129 and n = 92; GLU, n = 0 and n = 0; IP, n = 66 and n = 24. [Colour figure can be viewed at wileyonlinelibrary.com]

Extensor bursts started before paw contact with the treadmill and terminated before or at the end of the stance phase (Fig. 1). The burst onset times normalized to the cycle times were significantly affected by state (F _1,5479 = 106.1, P < 0.001), speed (F _2,5479 = 420.7, P < 0.001), and muscle (F _12,5479 = 207.4, P < 0.001). For example, the relative burst onset times occurred significantly later in the cycle in the spinal state compared to intact state at locomotor speeds of 0.4 m/s (0.29 ± 0.09 vs. 0.28 ± 0.06, P < 0.001), 0.7 m/s (0.35 ± 0.10 vs. 0.32 ± 0.07, P < 0.001) and 1.0 m/s (0.40 ± 0.10 vs. 0.35 ± 0.08, P < 0.001). (Note that the normalized swing duration is substantially longer in the spinal than in intact state (see below), which creates an appearance of an earlier burst onset in the spinal state.) The extensor burst onset times also occurred later in the cycle with increasing speed (F _2,5479 = 542.8, P < 0.001). The normalized extensor burst offset times significantly depended on the speed, muscle and state–speed–muscle interaction (F _2,5466 = 46.2, P < 0.001; F _12,5468 = 1073.1, P < 0.001; and F _60,5465 = 19.6, P < 0.001; respectively) but not on the state (F _1,5472 = 1.7, P = 0.187). The normalized offset times of extensor bursts occurred later in the cycle in the spinal state compared to intact state at speeds 0.4 m/s and 1.0 m/s (0.76 ± 0.12 vs. 0.73 ± 0.12, P < 0.001 and 0.77 ± 0.12 vs. 0.75 ± 0.12, P = 008, respectively) but not at 0.7 m/s (0.76 ± 0.12 vs. 0.75 ± 0.12, P = 790).

The main effect of state on the mean amplitude of flexor bursts was not significant (F _1,2035 = 2.427, P = 0.119), whereas speed, muscle and state‐speed‐muscle interactions significantly affected flexor burst magnitude (F _2,2035 = 82.1, P < 0.001; F _6,2035 = 78.6, P < 0.001; and F _28,2035 = 14.6, P < 0.001; respectively). The majority of significant differences in mean amplitude between states occurred at a speed of 1.0 m/s, as most flexor bursts were greater after spinal transection (P < 0.001), except for ST, which was smaller in the spinal state (P < 0.001), and for PLO whose magnitude was not significantly different (P = 0.709) (Fig. 2B ). At slower speeds, fewer differences in burst magnitude between states could be seen and they often had the opposite signs, for example, TA at speed 0.7 m/s (P < 0.001) and BFP at speed 0.4 m/s (P < 0.001). At the slowest speed of 0.4 m/s, only two flexor bursts differed between states (BFP and ST) and both were greater in the intact state. The GLU flexor burst appeared only in the intact state and at two speeds, 0.4 and 0.7 m/s (Figs 1 and 2B ).

The flexor bursts typically started prior to swing onset and terminated at or slightly after swing offset (Fig. 1). The normalized flexor burst onset times were dependent on the state, locomotion speed, muscle and their interaction (F _1,2013 = 160.9, P < 0.001; F _2,2018 = 3.4, P = 0.035; F _6,2005 = 59.3, P < 0.001; and F _28,2017 = 35.6, P < 0.001; respectively). Flexor bursts started prior to swing onset in the intact state at all speeds (0.4 m/s: −0.07 ± 0.04; 0.7 m/s: −0.60 ± 0.04; and 1.0 m/s: −0.07 ± 0.04), whereas in the spinal state the onset started later and closer in time to swing onset (0.4 m/s: −0.04 ± 0.06; 0.7 m/s: 0.003 ± 0.08; and 1.0 m/s: 0.01 ± 0.08). Flexor burst offset times were significantly affected by speed (F _2,2017 = 69.4, P < 0.001), muscle (F _6,2010 = 332.2, P < 0.001), and state–speed–muscle interaction (F _28,2017 = 61.6, P < 0.001), but not by state (F _1,2017 = 0.86, P = 0.855). Flexor burst offset times did not depend on state at speeds of 0.4 and 1.0 m/s (0.17 ± 0.09 vs. 0.21 ± 0.10, P = 0.539 and 0.23 ± 0.14 vs. 0.23 ± 0.12, P = 0.479, in the intact and spinal state, respectively); at 1.0 m/s, the flexor bursts terminated later in the cycle in the spinal state than in the intact state (0.25 ± 0.13 vs. 0.19 ± 0.12, P < 0.001).

Flexor and extensor burst normalized timings were also related to the durations of the stance and swing phases (Fig. 1). Figure 3 shows the durations of the cycle, stance and swing phases as well as the duty cycle and stride length for all cats and experimental conditions. The cycle time was shorter in the spinal state compared to the intact state at all three speeds (0.4 m/s: 0.768 ± 0.130 s vs. 1.070 ± 0.144 s, P < 0.001; 0.7 m/s: 0.656 ± 0.132 s vs. 0.792 ± 0.070 s, P < 0.001; 1.0 m/s: 0.572 ± 0.079 vs. 0.636 ± 0.051, P < 0.001) (Fig. 3A ). The effects of state, speed and their interaction were significant (F _1,2493 = 1476.9, P < 0.001; F _2,2493 = 1717.6, P < 0.001; F _2,2493 = 256.8, P < 0.001; respectively).

Figure 3. Bar plots and boxplots of kinematic characteristics of locomotion cycles of individual cats.

Bar plots with individual data points are shown for experimental conditions and cats with n < 30. Boxplots are shown when n ≥ 30. The boxes in boxplots show the range of values between 25th and 75th percentiles with the median inside the boxes, and whiskers indicating minimal and maximal values. Horizontal lines and P‐values above pairs of experimental conditions indicate significance of the difference between the conditions. A, cycle time. B, swing time. C, stance time. D, duty cycle. In panels A–D, number of data points for nine individual cats in intact condition at speed 0.4 m/s are: 20, 45, 28, 39, 50, 51, 46, 64, 31, respectively; in spinal condition at speed 0.4 m/s: 60, 29, 12, 44, 60, 51, 132, 59, 27; in intact condition at speed 0.7 m/s: 42, 59, 31, 53, 71, 15, 57, 79, 36; in spinal condition at speed 0.7 m/s: 51, 62, 15, 39, 69, 34, 73, 60, 65; in intact condition at speed 1.0 m/s: 29, 46, 27, 58, 76, 15, 12, 72, 30; in spinal condition at speed 1.0 m/s: 33, 28, 16, 38, 75, 40, 76, 40, 29. E, stride length. Number of data points for nine individual cats in intact condition at speed 0.4 m/s are: 14, 11, 13, 13, 11, 16, 14, 19, 15; in spinal condition at speed 0.4 m/s: 15, 11, 15, 15, 15, 15, 25, 10, 15; in intact condition at speed 0.7 m/s: 11, 15, 15, 15, 19, 16, 19, 19, 15; in spinal condition at speed 0.7 m/s: 12, 13, 15, 15, 19, 11, 19, 14, 15; in intact condition at speed 1.0 m/s: 14, 9, 14, 15, 19, 16, 12, 19, 10; in spinal condition at speed 1.0 m/s: 5, 12, 15, 12, 19, 12, 20, 7, 15. [Colour figure can be viewed at wileyonlinelibrary.com]

Swing duration was significantly affected by speed (F _1,2493 = 41.1, P < 0.001) and state–speed interaction (F _2,2493 = 122.7, P < 0.001) but not state alone (F _2,2493 = 2.5, P = 0.114). In the intact state, swing duration decreased with increasing speed from 0.325 ± 0.065 s at speed 0.4 m/s to 0.284 ± 0.034 s at speed 0.7 m/s (P < 0.001) and to 0.250 ± 0.027 s (P < 0.001) at speed 1.0 m/s (Fig. 3B ). In the spinal state, swing duration increased from 0.277 ± 0.077 s at 0.4 m/s to 0.299 ± 0.078 s at speed 0.7 m/s (P < 0.001). The swing duration at 1.0 m/s was not significantly different from 0.7 m/s (0.296 ± 0.059 s, P = 0.480).

State, speed and their interaction significantly affected stance duration (F _1,2493 = 2752.5, P < 0.001; F _2,2493 = 2573.0, P < 0.001; and F _2,2493 = 169.4, P < 0.001; respectively), which decreased with increasing speed in both states and was smaller in the spinal state compared to intact state (Fig. 3C ). Stance durations in the intact and spinal states were 0.745 ± 0.121 s and 0.491 ± 0.090 s at 0.4 m/s, respectively; 0.508 ± 0.058 s and 0.358 ± 0.089 s at 0.7 m/s; and 0.385 ± 0.40 s and 0.276 ± 0.60 s at 1.0 m/s. The duty cycle, the ratio of the stance duration over the cycle duration, was significantly affected by state, speed and their interaction (F _1,2493 = 1380.0, P < 0.001; F _2,2493 = 848.2, P < 0.001; and F _2,2493 = 67.8, P < 0.001; respectively). The duty cycle was significantly smaller in the spinal state compared to intact and decreased with increasing speed in both states (Fig. 3D ). The duty cycle values for intact and spinal states were 0.695 ± 0.049 and 0.642 ± 0.067, respectively, at 0.4 m/s; 0.642 ± 0.036 and 0.545 ± 0.079 at 0.7 m/s; and 0.606 ± 0.034 and 0.482 ± 0.078 at 1.0 m/s. A duty cycle below 0.5 indicates a running gait (Hildebrand, 1989). Thus, a walk‐to‐run transition occurred in spinal cats at a speed between 0.7 m/s and 1.0 m/s, while intact cats performed a walking gait in the entire range of studied speeds up to 1.0 m/s. In the spinal state at 1.0 m/s, the duty cycle was between 0.38 ± 0.05 and 0.47 ± 0.04 in five cats (running gait), 0.50 ± 0.09 in one cat (walk‐run transition), and between 0.51 ± 0.02 and 0.58 ± 0.03 in three cats (walking).

The hindlimb stride length was significantly affected by the state (F _1,778 = 150.8, P < 0.001), speed (F _2,778 = 869.1, P < 0.001) and their interaction (F _2,778 = 31.2.0, P < 0.001). Stride length was shorter in the spinal state compared to intact state, except at speed 1.0 m/s at which stride length did not differ statistically between spinal and intact states, and most spinal cats ran. Stride length also increased with locomotor speed (Fig. 3E ). It was 0.299 ± 0.046 m for spinal vs. 0.395 ± 0.056 m for intact at 0.4 m/s (P < 0.001), 0.452 ± 0.073 m vs. 0.497 ± 0.045 m at 0.7 m/s (P < 0.001), and 0.557 ± 0.80 m vs. 0.571 ± 0.48 m at 1.0 m/s (P = 0.062).

The general kinematic characteristics of the locomotor cycle described above are consistent with previous studies of intact and spinal cats (Frigon et al., 2014; Frigon et al., 2017; Halbertsma, 1983), including a decrease in cycle and stance durations with increasing speed and an earlier walk‐to‐run transition in spinal cats.

EMG burst groups

We used the minimum spanning tree clustering algorithm in MATLAB to determine synergistic groups of 15 muscles that are activated together, that is, whose EMG bursts occur in the same phase of the locomotion cycle (see Methods). Based on the obtained burst clusters and previous similar EMG burst group analyses (Klishko et al., 2021; Krouchev et al., 2006; Markin et al., 2012), we identified four or five major EMG burst groups and their subgroups in each experimental condition (Figs 4, 5, 6). Group 1 consisted of flexor bursts of BFP and ST, except in spinal locomotion at 0.4 m/s and 1.0 m/s where these bursts were combined by the clustering algorithm with the bursts of flexors TA, SRTa and IP (Figs 4B , 6B ). Group 1 was located at the stance–swing transition on the burst onset–offset time plots (see also Fig. 1). Group 2 consisted of bursts of flexors TA, STRa and IP (subgroup 2₁) and flexor bursts of GLU_F and PLO_F (subgroup 2₂) when present, with all these bursts occurring during swing. During intact locomotion at 1.0 m/s, subgroup 2₁ was separated into smaller subgroups TA and SRTa_F‐IP (Fig. 6A ). All group 2 flexor bursts occurred during swing (Fig. 1). Burst group 3 consisted of extensor bursts of BFP_E and ST_E (in spinal state at speed 0.4 m/s ST_E burst was missing) and occurred at the swing–stance transition. The burst group 4 combined EMG bursts of all hindlimb extensors and extensor bursts of bifunctional PLO and GLU. In some experimental conditions (intact at 0.4 and 1.0 m/s and spinal at 0.7 m/s), bursts of single muscles (BFA and CF, Fig. 4A ; FDL, Fig. 5B ; PLO and VL, Fig. 6A ) were separated into distinct subgroups. In spinal cats at speeds of 0.7 and 1.0 m/s, burst group 4 was combined with extensor group 5 (Figs 5B and 6B ). Group 5 consisted of the SRTa extensor burst, which occurred in all experimental conditions except one (intact at 0.4 m/s), but was clustered separately from group 4 in spinal locomotion at 0.4 m/s and intact locomotion at 0.7 and 1.0 m/s. Burst group 5 occurred late in stance and was similar to burst group 5 identified previously during level and upslope overground walking (Klishko et al., 2021) and included rectus femoris (knee extensor, hip flexor).

Figure 4. Scatter plots of EMG burst onset and offset times of 15 hindlimb muscles and EMG burst groups for locomotion speed of 0.4 m/s in intact (A) and spinal (B) states.

Onset and offset times are shown as cycle phase from 0 (swing phase onset) to 1.0 (next swing phase onset). Ellipses and corresponding colour‐coded symbols denote EMG burst groups/subgroups identified by cluster analysis. The grey dashed lines indicate the onset of the flexor and extensor phases defined by the mean EMG burst onset time minus one SD of IP and SO, respectively. The blue dashed‐dotted lines separate the swing and stance phases. For muscle abbreviations see the legend for Fig. 1. [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 5. Scatter plots of EMG burst onset and offset times of 15 hindlimb muscles and EMG burst groups for locomotion speed of 0.7 m/s in intact (A) and spinal (B) states.

For further information see the legend for Fig. 4. [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 6. Scatter plots of EMG burst onset and offset times of 15 hindlimb muscles and EMG burst groups for locomotion speed of 1.0 m/s in intact (A) and spinal (B) states.

For further information see the legend for Fig. 4. [Colour figure can be viewed at wileyonlinelibrary.com]

Muscle synergies and their activation patterns determined by non‐negative matrix factorization

Muscle synergies

EMG burst groups identified based on burst onset and offset times indicate synergistic muscles that are active at the same time, consistent with receiving a common neural input (Krouchev et al., 2006; Markin et al., 2012). Contributions of individual group members to the total EMG activity of the group and group activation patterns can be further assessed by a muscle synergy analysis using NNMF (Cheung et al., 2005; Klishko et al., 2021). We first evaluated the number of muscle synergies that accounted for at least 90% of EMG variance in all muscles in each experimental condition. This number was found to be five in all experimental conditions (Fig. 7A ). The main effect of state on the EMG variance accounted for (the coefficient determination R ² computed between the recorded and reconstructed from synergies EMG patterns) was insignificant (F _1,343 = 2.6, P = 0.105), whereas the effects of speed and state‐speed interaction were significant (F _2,343 = 8.7, P < 0.001 and F _2,343 = 15.6, P < 0.001, respectively). Although significant, the differences in R ² between the experimental conditions were rather small: the minimum and maximum values were 0.91 ± 0.02 (intact at 0.4 m/s) and 0.93 ± 0.02 (intact at 1.0 m/s). To evaluate the similarity of matrices W and C in different experimental conditions, we also computed the variance accounted for by different combinations of matrices D^* _COM = W _COM C _COM which were composed by mixing different experimental conditions (see Methods). As seen in Fig. 7B , five synergies were insufficient to obtain an R ² at or above 0.9 for any combination of experimental conditions. Seven synergies were necessary to obtain R ² > 0.9 for the combination intact‐spinal state at each speed, that is, ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4}^{\ast }$, ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.7}^{\ast }$ and ${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_1.0}^{\ast }$ (first three bars for seven synergies) and eight synergies were required for an R ² to exceed 0.9 by the combination of all speeds for intact and spinal states, that is, ${\mathbf{D}}_{\mathrm{In}\_0.4|0.7|1.0}^{\ast }$ and ${\mathbf{D}}_{\mathrm{Sp}\_0.4|0.7|1.0}^{\ast }$ (next two bars). The combination of all six experimental conditions (${\mathbf{D}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}^{\ast }$) required 10 synergies to reach an R ² of 0.9 (Fig. 7B ). These results suggest that the five muscle synergies and/or their activation patterns found for each experimental condition differ between the conditions. The fact that the combination of all six experimental conditions required the greatest number of synergies (n = 10) to account for 90% of variance in the complete set of experimental EMG patterns suggests that each experimental condition introduces a unique contribution to the synergy formation.

Figure 7. Variance accounted for (mean ± SD) in recorded EMG patterns by EMG patterns reconstructed from muscle synergies as a function of the number of synergies.

A, variance accounted for by reconstructed EMG patterns in each intact and spinal state during locomotion at 0.4, 0.7 and 1.0 m/s. Horizontal lines indicate values of variance 0.85, 0.90 and 0.95. B, variance accounted for by reconstructed EMG patterns obtained from different combinations of matrices D^* _COM = W _COM C _COM composed by mixing different experimental conditions (see Methods). Variance computed for six combinations is shown: bars 1–3, the intact and spinal state combination at speeds 0.4, 0.7 and 1.0 m/s, respectively; bars 4–5, the combination of three speeds for intact and spinal state, respectively; bar 6, the combination of all six experimental conditions. [Colour figure can be viewed at wileyonlinelibrary.com]

Five muscle synergies (coefficients W) and their activation patterns (coefficients C) are shown in Fig. 8A and B , respectively, for each experimental condition and for the combination of all experimental conditions together. The order of muscle synergies was selected based on the appearance of the synergy's activation peak in the locomotor cycle. Thus, the first two synergies had activity peaks at the onset and middle of swing, respectively, in all experimental conditions and in the combination of all conditions (Fig. 8B ). The activation peak of synergy 3 occurred near the swing–stance transition, the activation peak of synergy 4 at early stance; and the activation peak of synergy 5 at middle‐late stance in all experimental conditions. Muscle contributions to the corresponding synergies (coefficients W) are shown in Fig. 8A . Although the median of most coefficients W in all synergies and experimental conditions exceeded zero (non‐parametric Wilcoxon signed‐rank test, P < 0.001, n = 50), that is, each muscle made a non‐zero contribution to all synergies, we considered the weight coefficients W below 0.15 negligible. The value of 0.15 was selected arbitrary because it approximately corresponded to the maximum range of jitter of coefficients W near zero. The linear mixed‐effect model analysis performed for individual synergies revealed significant effects on coefficients W of the factor muscle in all synergies (F _14,8820 = 252.3–1161.4, P < 0.001), state in synergies 3 and 4 (F _14,8820 = 252.3, P <0.001 and F _14,8820 = 327.8, P < 0.001) but not synergies 1, 2 and 5 (F _14,8820 = 1.8–2.4, P = 0.089–0.157), and speed in synergies 1, 3–5 (F _3,8820 = 4.4–22.2, P ≤ 0.004) but not synergy 2 (F _3,8820 = 1.5, P = 0.219).

Figure 8. Five muscle synergies and their activation patterns during tied‐belt locomotion at speeds 0.4, 0.7 and 1.0 m/s in intact and spinal states.

A, mean (±SD) muscle synergy weights (relative contribution of each muscle to a given synergy; columns in (15 $\times$ 5) matrices ${\mathbf{W}}_{\mathrm{In}\_0.4}$, ${\mathbf{W}}_{\mathrm{In}\_0.7},{\mathbf{W}}_{\mathrm{In}\_1.0},{\mathbf{W}}_{\mathrm{Sp}\_0.4}$, ${\mathbf{W}}_{\mathrm{Sp}\_0.7},{\mathbf{W}}_{\mathrm{Sp}\_1.0}$ corresponding to each experimental condition: intact and spinal states at three locomotion speeds). Matrix ${\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}$ contains muscle weight coefficient for a combination of all 6 experimental conditions; matrix dimensions are (15 $\times$ 5). Each matrix W was computed using 50 randomly selected cycles across all cats (see Methods). B, mean (±SD) activation patterns of five muscle synergies (rows in (5 $\times$ 101) matrices ${\mathbf{C}}_{\mathrm{In}\_0.4}$, ${\mathbf{C}}_{\mathrm{In}\_0.7},{\mathbf{C}}_{\mathrm{In}\_1.0},{\mathbf{C}}_{\mathrm{Sp}\_0.4}$, ${\mathbf{C}}_{\mathrm{Sp}\_0.7},{\mathbf{C}}_{\mathrm{Sp}\_1.0}$ corresponding to each experimental condition)0. The activation patterns for each experimental condition are shown as a function of the normalized cycle time in the following order: intact (int), speed 0.4 m/s; intact, speed 0.7 m/s; intact, speed 1.0 m/s; spinal (sp), speed 0.4 m/s; spinal, speed 0.7 m/s; and spinal, speed 1.0 m/s. Activation coefficients in matrix ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}$ (orange line) represent synergy activation patterns of a combination of all 6 experimental conditions; matrix dimensions are (5 $\times$ 606). Vertical continues lines surrounded by vertical dashed lines correspond to the mean ± SD swing offset/stance onset normalized times in each experimental condition. These lines separate the swing (sw) and stance (st) phases. For muscle abbreviations see the legend for Fig. 1. [Colour figure can be viewed at wileyonlinelibrary.com]

Bifunctional muscles BFP and ST made the greatest contributions to synergy 1 during intact locomotion at all speeds (range 0.68–0.85) with much smaller contributions of flexors TA, SRTa and IP (range 0.12–0.35). In spinal locomotion, these contributions were significantly smaller for BFP and ST (below 0.15 at speeds 0.4 and 0.7 m/s; 0.31–0.35 at 1.0 m/s), but larger for TA, SRTa and IP at all speeds (0.28–0.68) except for TA at speed 0.4 m/s (0.13). The contributions of extensor muscles or flexor bursts of bifunctional muscles to synergy 1 was negligible (below 0.15) except for a small PLO contribution in intact locomotion at 1.0 m/s (0.19) and spinal locomotion at 0.4 m/s (0.16).

In synergy 2, flexors TA, SRTa and IP significantly increased their contributions in the intact and spinal states at all three speeds (coefficients W ranged from 0.44 to 0.83, P ≤ 0.014) except for SRTa in spinal locomotion at speed 0.7 m/s (P = 0.868) and IP in spinal locomotion at speed 0.4 m/s (P = 0.186). The contributions of the bifunctional BFP and ST were negligible in intact locomotion at all speeds but significantly increased for BFP in spinal locomotion (range 0.38–0.45) compared to its contribution to synergy 1 at 0.4 and 0.7 m/s (P < 0.001) but not at 1.0 m/s (P = 0.421). The ST contribution to synergy 2 in spinal locomotion was small (0.10–0.25). The contribution of the PLO extensor burst was also small at all speeds in intact (range 0.19–0.22) and spinal (range 0.04–0.08) locomotion. The contribution of the extensor burst of GLU exceeded 0.15 only in intact locomotion at 0.4 m/s (0.21). The W coefficients of extensor muscles were below 0.15 and thus did not contribute substantially to synergy 2. Major contributors to synergies 1 and 2 (BFP, ST, TA, SRTa and IP) corresponded to EMG burst groups 1, 2 and 3 (Figs 4, 5, 6).

All extensor muscles contributed substantially to synergies 3, 4 and 5 during intact and spinal locomotion at all speeds. The greatest contributions to synergies 3 and 4 were provided by ankle extensors SO, MG, LG and PL in intact (range 0.38–0.63) and spinal (0.21–0.54) locomotion (with significantly higher W coefficients in intact than spinal locomotion, P ≤ 0.05) with very small number of exceptions. For example, the change was insignificant between the conditions for MG, synergy 3 at 1.0 m/s and synergy 4 at 0.7 m/s, P = 0.203–0.714; SO, synergy 4, speed 0.4 m/s, P = 0.458; and PL, synergy 4, speed 0.7 m/s, P = 0.105. The greatest contributions to extensor synergy 5 in all experimental conditions were provided by proximal muscles VL (range 0.542–0.896), BFA (0.248–0.519), GLU (0.152–0.535) and CF (0.452–0.785). Distal extensors contributed less to synergy 5; in intact locomotion their contribution was between 0.125 (LG, speed 0.7 m/s) and 0.304 (SO, speed 0.4 m/s). The contributions of flexors TA and IP and extensor bursts of SRTa, BFP and ST to extensor synergies were substantial only in synergy 3, primarily in spinal locomotion, which ranged between 0.201 in SRTa at 0.4 m/s and 0.402 in TA at 0.4 m/s. The contributors to extensor synergies 3–5 corresponded to EMG burst groups 4 and 5 (Figs 4, 5, 6).

When we compared muscle weight coefficients W between different combinations of states and speeds (various matrices W _COM) to evaluate similarity of synergies across the experimental conditions, we found generally similar distributions of muscle contributions to individual synergies compared to those revealed by synergy analysis of individual experimental conditions (compare Fig. 8A and Fig. 9). Specifically, flexor bursts of bifunctional BFP and ST made the greatest contributions to synergy 1 in the majority of state and speed combinations except for spinal locomotion at all speeds and the combination of intact and spinal locomotion at 0.7 m/s (Fig. 9). Flexor muscles TA, SRTa and IP made generally lower contributions to synergy 1 but were main contributors to synergy 2 (typically above 0.5 and reaching a peak value of 0.922 ± 0.125 for SRTa in intact locomotion at all speeds). In extensor synergies 3–5, flexors and bifunctional BFP and ST made negligible contributions (in most cases below 0.15), whereas extensors and extensor bursts of PLO and GLU contributed substantially. As was the case in the individual experimental conditions (Fig. 8A ), distal ankle extensors contributed more to synergies 3 and 4 (except for GLU whose contributions were between 0.359 and 0.837 for 4 out of 6 experimental combinations), while proximal extensors VL, BFA, GLU and CF contributed to synergy 5 with weights typically exceeding 0.5 (Fig. 9).

Figure 9. Five muscle synergies (mean ± SD) computed for different matrices W_COM composed by mixing different combinations of experimental conditions (see Methods).

These combinations are (in the order of bars for each muscle): (1) the combination of all 6 experiments conditions (intact, spinal, speeds 0.4, 0.7 and 1.0 m/s), matrix ${\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.4|0.7|1.0}$; (2) intact and the combination of all speeds, ${\mathbf{W}}_{\mathrm{In}\_0.4|0.7|1.0}$; (3) spinal and the combination of all speeds, ${\mathbf{W}}_{\mathrm{Sp}\_0.4|0.7|1.0}$; (4) the combination of intact and spinal state at speed 0.4 m/s, ${\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$; (5) the combination of intact and spinal state at speed 0.7 m/s, ${\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$; and (6) the combination of intact and spinal state at speed 1.0 m/s, ${\mathbf{W}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$. For muscle abbreviations see the legend for Fig. 1. [Colour figure can be viewed at wileyonlinelibrary.com]

In summary, we found that five synergies were sufficient to account for at least 90% of the variance in EMG patterns in each of six experimental conditions. Although in all experimental conditions the first two synergies were flexor and the remaining three synergies extensor, the composition (muscle weights) of the five synergies significantly differed among the intact and spinal states and locomotion speeds.

Synergy activation patterns

Activation patterns (time‐dependent coefficients C, Fig. 8B ) of all synergies were significantly affected by the factors time (F _9,11 760 = 709.2–2023.4, P < 0.001), state (F _3,11 760 = 29.2–97.6, P < 0.001) and speed (F _5,11 760 = 30.1–86.4, P < 0.001). In synergy 1, activation patterns of intact locomotion at different speeds were highly correlated, with mean coefficients of determination (R ²) computed between all pairs of C coefficient vectors corresponding to different speeds between 0.68 and 0.69. These patterns were also highly correlated with the corresponding components of common matrix ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$: R ² = 0.90–0.93 (Fig. 10A , orange lines, see Methods). The intact activation patterns at different speeds had a relatively low correlation with the corresponding spinal patterns (R ² = 0.30–0.49). In spinal locomotion, the activation patterns were dissimilar at different speeds as indicated by low R ² coefficients (0.27–0.28) and their correlation with the corresponding components of common matrix ${\mathbf{C}}_{\mathrm{Sp}\_0.4|0.7|1.0}$ was low: R ² = 0.21–0.37 (Fig. 10B ). Components of common matrices ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$, combining intact and spinal states at each speed, demonstrated high correlation only with the intact matrices ${\mathbf{C}}_{\mathrm{In}\_0.4}$ and ${\mathbf{C}}_{\mathrm{In}\_1.0}$ at 0.4 m/s (R ² = 0.90) and 1.0 m/s (R ² = 0.91) in synergy 1 (Fig. 10C and E , synergy 1). The spinal components of the above matrix C combining intact and spinal states demonstrated much lower correlations with the C coefficients for spinal locomotion at different speeds (R ² = 0.23–0.49, Fig. 10C–E ).

Figure 10. Synergy activation patterns (mean ± SD) as a function of the normalized cycle time for individual experimental conditions and their combinations.

A, activation patterns for intact state and different locomotion speeds are shown in the following order from left to right: speed 0.4 m/s, speed 0.7 m/s and speed 1.0 m/s. Activation coefficients in the common matrix ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$ (orange line) represent synergy activation patterns of a combination of all three speeds of intact state; the matrix dimensions are (5 $\times$ 303). Vertical continues lines surrounded by vertical dashed lines correspond to the mean ± SD swing offset/stance onset normalized times in each experimental condition. These lines separate the swing (sw) and stance (st) phases. Coefficients of determination (R ²) indicate correlation between the activation pattern of an individual experimental condition with the corresponding component of the common matrix ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$. B, activation patterns for spinal state and different locomotion speeds are shown in the following order from left to right: speed 0.4 m/s, speed 0.7 m/s and speed 1.0 m/s. Activation coefficients in the common matrix ${\mathbf{C}}_{\mathrm{Sp}\_0.4|0.7|1.0}$ (orange line) represent synergy activation patterns of a combination of all three speeds of spinal state. Coefficients of determination (R ²) indicate correlation between the activation pattern of an individual experimental condition with the corresponding component of the common matrix ${\mathbf{C}}_{\mathrm{Sp}\_0.4|0.7|1.0}$. C, activation patterns for intact and spinal states at locomotion speed of 0.4 m/s and the activation coefficients in the common matrix ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$ (orange line). D, activation patterns for intact and spinal states at locomotion speed of 0.7 m/s and the activation coefficients in the common matrix ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$ (orange line). D, activation patterns for intact and spinal states at locomotion speed of 1.0 m/s and the activation coefficients in the common matrix ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$ (orange line). [Colour figure can be viewed at wileyonlinelibrary.com]

Activation patterns of synergy 2 were generally consistent within and between states and speeds. Correlations for intact locomotion between speeds were R ² = 0.65–0.71 and R ² = 0.37–0.51 for spinal locomotion. Intact patterns at different speeds had a modest correlation with the corresponding spinal patterns (R ² = 0.47–0.60). Correlations of components of common matrix ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$ corresponding to different speeds were high with the corresponding matrices ${\mathbf{C}}_{\mathrm{In}\_0.4}$, ${\mathbf{C}}_{\mathrm{In}\_0.7}$, and ${\mathbf{C}}_{\mathrm{In}\_1.0}$ of intact locomotion (R ² = 0.83–0.92); similar correlations for spinal locomotion were lower (R ² = 0.50–0.62) (Fig. 10A and B ). Correlations of components of common matrices ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$ with the patterns for intact locomotion were medium–high at 0.4 and 1.0 m/s (R ² = 0.43–0.86) and low for speed 0.7 m/s (R ² = 0.24–0.35) (Fig. 10C–E ).

Activation patterns of extensor synergies 3–5 corresponding to different speeds were modestly correlated in intact locomotion (R ² = 0.43–0.66) and had a wide range of correlations in spinal locomotion from very low to modest (R ² = 0.08–0.63). Correlations between intact and spinal patterns at the same speeds were low to modest (R ² = 0.21–0.58). Correlations of components of common matrices ${\mathbf{C}}_{\mathrm{In}\_0.4|0.7|1.0}$ and ${\mathbf{C}}_{\mathrm{Sp}\_0.4|0.7|1.0}$ with the corresponding matrices C for individual states and speeds were much higher for intact locomotion (R ² = 0.46–0.79) than for spinal locomotion (R ² = 0.18–0.61) (Fig. 10A and B ). Correlations of components of common matrices ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.4}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_0.7}$, ${\mathbf{C}}_{\mathrm{In}|\mathrm{Sp}\_1.0}$ with the corresponding individual matrices C were modest–high for intact locomotion (R ² = 0.44–0.84) and low–modest for spinal locomotion (R ² = 0.23–0.67) at all speeds (Fig. 10C–E ).

In summary, activation patterns of each synergy in intact locomotion at different speeds were similar to each other and to the components of the common matrices C _COM with very few exceptions. Activation patterns of spinal locomotion, on the other hand, were mostly different across speeds and dissimilar compared to the components of the common matrices C _COM. Nevertheless, the phases of high activation and the appearance of the activation peaks in the cycle were consistent across states and speeds (Figs 8B and 10).

Discussion

The goal of this study was to determine hindlimb muscle synergies and their activation patterns during walking at different locomotion speeds before and after spinal transection. The obtained results supported our hypothesis that a spinal mechanism controls the number of muscle synergies as a function of speed during locomotion. We found that five synergies were sufficient to account for 90% of the EMG variance in each of six experimental conditions (i.e. 2 states × 3 speeds). However, the obtained muscle contributions to each synergy (weights W) and the synergy activation patterns (time‐dependent coefficients C) in the different experimental conditions did not support our hypothesis that a spinal mechanism fully explains the composition and activation patterns of muscle synergies during locomotion at different speeds.

State‐dependent effects on muscle synergies

Previous studies of EMG burst groups and muscle synergies during tied‐belt treadmill locomotion after low‐thoracic spinal transection demonstrated that hindlimb muscle synergies in adult cats originate in the lumbar spinal cord (Desrochers et al., 2019; Harnie et al., 2021; Higgin et al., 2020) and are generally similar to the EMG burst groups and synergies during intact cat locomotion (Klishko et al., 2021; Krouchev et al., 2006; Markin et al., 2012). The results of the present study are generally consistent with previous conclusions. We also found between five and nine EMG burst subgroups, which we combined in five major groups (Figs 4, 5, 6), consistent with groups/subgroups previously identified in both intact and spinal cats during forward locomotion. In the present study, spinal transection reduced the number of subgroups from seven to five at a speed of 0.4 m/s by merging group 1 and subgroup 2₁ (Fig. 4), from seven to six at 0.7 m/s by merging subgroup 4₂ and group 5 (Fig. 5), and from nine to four at 1.0 m/s by merging group 1 with subgroup 2₁ and subgroup 4₂ with group 5 (Fig. 6). These results are consistent with a previous study (Higgin et al., 2020) reporting a decrease in the number of burst groups from five to two during level intact and spinal locomotion at 0.4 m/s, respectively. Another study (Desrochers et al., 2019) also demonstrated a more compact grouping of muscle bursts that included SRTa and tensor fascia latae after spinal transection due to changes in burst onsets/offsets in these muscles.

We also found a substantial shift in the normalized EMG onset time in extensors around mid‐swing (Fig. 1). This shift in the flexor–extensor phase transition after spinal transection could result from the state‐machine operation regime of spinal networks that depends on motion‐dependent sensory feedback from the hindlimbs for phase transitions (Rybak et al., 2024). Indeed, previous studies of hindlimb‐only treadmill locomotion in spinal cats demonstrated a maximum hip flexion much earlier in the swing phase, closer to mid‐swing, than in intact cats (see Fig. 3 in Harnie et al. (2022)). Maximum hip flexion corresponds to the maximum length of the hamstring muscles (Gregor et al., 2006; Klishko et al., 2021) and peak activity of group I and II muscle spindle afferents from these muscles (Prochazka & Gorassini, 1998). The occurrence of peak hip flexion and hamstring length in the swing phase is strongly correlated with the onset of hindlimb extensor activity (Gregor et al., 2006; McVea et al., 2005).

The general composition of the muscle synergies (synergies 1 and 2 and 3–5 for flexors and extensors, respectively) and their activation patterns were qualitatively similar in intact and spinal locomotion (Figs 8, 9, 10). However, more in‐depth analyses (statistical tests and matrices W _COM and C _COM) revealed significant differences in both synergy composition and activation patterns (Figs 9 and 10). Although the exact reasons for and mechanisms responsible for these differences cannot be deduced from our data, we speculate that they could potentially result from well‐documented changes in spinal neurons (including membrane receptor and neurotransmitter production, synaptic connections, dendritic sprouting, etc.) and in afferent and propriospinal pathways due to spinal plasticity after spinal injury (Bilchak et al., 2021; Martin, 2022; Murray et al., 2010; Punjani et al., 2023; Rossignol & Frigon, 2011; Skinnider et al., 2024). Changes in hindlimb locomotor mechanics and thus in motion‐dependent somatosensory feedback also undoubtedly affect the composition and/or activation patterns of muscle synergies (Santuz et al., 2019).

Muscle synergies and their activation patterns during spinal locomotion in this study are much more consistent with intact condition than typically reported in studies of people with spinal cord injury (Danner et al., 2015; Fox et al., 2013; Hayes et al., 2014; Sun et al., 2022). This may be related to the fact that in human studies intact and spinal injury states cannot be compared in the same subjects, and it is difficult to recruit patients with the same location/severity of spinal injury, rehabilitation interventions, general health status, etc. It is also easier to control the posture of spinal cats when they step on the treadmill. Nonetheless, several features of muscle synergies in spinal cats and people with spinal cord injury appear common and include flexor and extensor synergies, and the inclusion in individual synergies muscles acting across several joints.

Speed‐dependent effects on muscle synergies

Locomotor speed did not affect general features of muscle EMG burst groups (Figs 4, 5, 6) and muscle synergies (number, general composition and pattern, Figs 7A and 8). However, increasing speed led to an increase in the normalized magnitude of extensor and flexor EMG bursts (Fig. 2), changes in the composition of muscle burst subgroups within generally the same burst groups (Figs 4, 5, 6), changes in muscle synergy weights (Fig. 8A ), and shifting phases and increasing peak magnitude of synergy activation patterns (Fig. 8B ) in both intact and spinal locomotion. All these changes are expected given that increasing speed results in an increased supraspinal drive and motion‐dependent somatosensory feedback in intact locomotion and an increased somatosensory feedback in spinal locomotion (Caggiano et al., 2018; Forssberg et al., 1980; Frigon et al., 2017; Shik et al., 1966).

Our results are consistent with human studies that demonstrated small variability in the number and composition of muscle synergies, with greater changes in synergy activation patterns within the same gait, walking or running (Dewolf et al., 2019; Gui & Zhang, 2016; Ivanenko et al., 2003; Monaco et al., 2010; Saito et al., 2018; Yokoyama et al., 2016). Substantial differences in muscle synergy composition and/or their activation patterns have been reported between walking and running in humans (Hagio et al., 2015; Ivanenko et al., 2008; Yokoyama et al., 2016). Five out of nine cats in spinal locomotion switched to a running gait at speed 1.0 m/s in our study. Although we did not notice abrupt changes in muscle synergies when this occurred, we did find more flexor EMG bursts with greater magnitude than in intact locomotion at 1.0 m/s, where a walking gait was maintained (Fig. 2B ). This result is consistent with previous observations that increased demands on leg flexors during the swing phase trigger the walk–run transition (Hreljac et al., 2007; Ivanenko et al., 2008; Prilutsky & Gregor, 2001).

Potential organization of the spinal locomotor network controlling a single hindlimb

The number and composition of muscle burst groups and synergies obtained in this study provide information on the possible organization of the spinal locomotor network controlling hindlimb muscle activity. Two flexor and three extensor synergies that include muscles spanning ankle, knee and hip joints appear consistent with a two‐level CPG organization proposed previously (McCrea & Rybak, 2008; Rybak, Shevtsova et al., 2006). The pattern formation (PF) level of such a model can contain five premotor interneuronal populations forming five muscle synergies. Two of these populations receive excitatory inputs from the flexor rhythm generator (RG) half‐centre and motion‐dependent somatosensory feedback, while the other 3 PF populations receive excitatory inputs from the extensor RG half‐centre and motion‐dependent somatosensory feedback (Fig. 11). According to the two‐level CPG organization, the somatosensory feedback received by PF interneuronal populations modulates the duration and magnitude of activity in flexor and extensor motoneurons without changing or resetting the locomotor cycle duration (non‐resetting feedback). Changes in locomotor rhythm can be accomplished by somatosensory feedback received by the RG half‐centres (resetting feedback, Fig. 11). The reciprocal inhibition between the PF flexor and extensor populations reflects a lack of coactivation between the activation patterns of flexor synergies 1 and 2 and extensor synergies 3–5 (Fig. 8B ). Reflex pathways with Ia inhibitory interneurons (not shown) provide additional contribution to reciprocal activity of flexors and extensors (Feldman & Orlovsky, 1975; Geertsen et al., 2011). The PF flexor and extensor populations provide excitatory inputs to motoneurons of corresponding flexor and extensor muscles with different weights. Motoneurons controlling bifunctional muscles with both flexion and extension actions at different joints receive excitatory inputs from both the PF flexor and extensor populations as suggested previously (Perret & Cabelguen, 1980; Prilutsky, 2000; Shevtsova et al., 2016). The preliminary computer simulations using our neuromechanical model of cat hindlimb spinal locomotion utilizing a similar organization of the CPG and somatosensory afferent inputs (Markin et al., 2016; Rahmati et al., 2023) reproduced basic kinematic and muscle activity patterns recorded experimentally.

Figure 11. Schematic representation of a spinal locomotor CPG for a single hindlimb consistent with five synergies revealed in spinal state.

The schematic diagram is a modified version of a previous two‐level CPG model (McCrea & Rybak, 2008; Rybak, Stecina et al., 2006). The model consists of a rhythm generator with flexor and extensor half‐centres (large circles) mutually inhibiting each other, and a pattern formation network that contains 5 premotor interneuronal populations (smaller circles) activating motoneurons (diamonds) of the corresponding muscles (ovals). For muscle abbreviations see the legend for Fig. 1. Additional muscle abbreviations: RF, rectus femoris (hip flexor, knee extensor) and SRTm, sartorius medial (hip flexor and knee flexor). These muscles were not investigated in this study and included in the schematic based on previous synergy analyses (Higgin et al., 2020; Klishko et al., 2021; Markin et al., 2012). See text for further explanations. [Colour figure can be viewed at wileyonlinelibrary.com]

Conclusions

To better understand the organization of spinal locomotor networks, we investigated the effects of low thoracic spinal cord transection and locomotion speed on EMG activity, EMG burst groups and muscle synergies in cats during treadmill locomotion at different speeds before and after spinal transection. We found that both the spinal transection and locomotion speed changed the locomotor cycle and phase durations, stride length, the EMG burst magnitude of hindlimb flexors and extensors, and the normalized EMG burst onset and offset times. The major EMG burst groups, the number of muscle synergies, and major features of muscle synergy composition and activation patterns were not substantially affected by spinal transection and locomotion speed, consistent with a spinal control mechanism. In all experimental conditions, we typically observed (i) five major EMG burst groups of hindlimb flexors and extensors; (ii) two muscle synergies composed of BFP and/or ST bifunctional muscles and flexors active at the stance–swing transition and during the swing phase, respectively; and (iii) three extensor synergies composed of hindlimb extensors and active at the swing–stance transition, the early stance phase and the mid–late stance phase, respectively. Both spinal transection and locomotion speed modified subgroups of the EMG burst groups and the composition and activation patterns of selected synergies. The obtained results suggest that supraspinal drive, plastic changes in the spinal cord after its transection and motion‐dependent sensory feedback do not modify the general organization of the spinal locomotor networks but rather produce more subtle changes possibly in neuronal properties, synaptic connections, and somatosensory and propriospinal pathways. Based on the obtained results, we proposed an organization of a PF network of a two‐level CPG, which can be tested in neuromechanical simulations of cat spinal locomotion.

Competing interests

The authors declare no competing interests.

Author contributions

A.F., B.I.P. and I.A.R. conceived and designed research. J.H. and A.F. performed surgeries and data collection. A.N.K. developed computer code. A.N.K., C.E.H. and S.M.R performed data analysis. A.N.K. and B.I.P. prepared figures. B.I.P. drafted manuscript. All authors discussed, reviewed, edited, and approved the final version of the manuscript. All authors have agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

This study was supported by US NIH grant R01 NS110550.

Supporting information

Peer Review History

Biography

Alexander N. Klishko obtained his Bachelor's and Masters's degrees in Applied Mathematics and Physics at the Moscow Institute of Physics and Technology, Moscow, Russia and PhD in Biophysics at the Institute of Theoretical and Experimental Biophysics, Pushchino, Russia. He completed his postdoctoral training in neural control of movement in the School of Applied Physiology (currently Biological Sciences) at Georgia Institute of Technology, Atlanta, GA, USA where he is currently a Senior Research Scientist. His general research interests are biomechanics and motor control of movement. He uses experimental electrophysiological and biomechanical methods and computational modelling to investigate how different animal species, including humans, control their motor behaviours in conditions of health and pathology.

Data availability statement

Low‐pass filtered EMG patterns obtained in this study will be made available on request. MATLAB code for computing all variables presented in the article is available at https://github.com/BorisPrilutskyLab/MuscleSynergies.