Differences in backward and forward treadmill locomotion in decerebrated cats

Despite the rapid development of medicine, there is still an urgent need for new neurorehabilitation approaches for the treatment of motor dysfunction. Thus, more attention needs to be focused on the mechanisms of both locomotor control per se and the functional processes underlying rehabilitation techniques. One of the techniques used in the treatment of multiple neurological disorders, especially in sports medicine, is backward body propulsion (walking, running and pedalling) (Hoogkamer et al., 2014; Balasukumaran et al., 2018). Backward movement initially attracted attention because of its ability to protect the knee joint from overload while simultaneously increasing the quadriceps strength and power (Flynn and Soutas-Little, 1993). Currently, it is widely used in multiple applications, such as stroke rehabilitation (Rose et al., 2018), cerebral palsy (Kim et al., 2013) and Parkinson's disease (Grobbelaar et al., 2017) therapy, and the total improvement of coordination and muscular strength (Flynn and Soutas-Little, 1993; Wang et al., 2019).

Locomotion itself is under the control of cerebral and spinal neuronal networks. The spinal neuronal core comprises so-called central pattern generators (CPGs) located within the cervical and lumbosacral spinal cord (Jankowska, 2008; Kiehn, 2016; Grillner and El Manira, 2020). In some studies, it has been proposed that CPGs consist of multiple levels including one for rhythm generation and another for pattern formation (McCrea and Rybak, 2008). Most views on CPGs and locomotor networks are generally based on forward locomotion. Forward locomotion is the most frequently used locomotor mode for most vertebrates (especially mammals) because it corresponds to the polarity of the head where several key sensors, such as the eyes, are located. Animals move their heads toward the source of visual, acoustic and odorant information; thus, forward locomotion is in line with the direction of gaze fixation, etc. In contrast, backward stepping is mostly associated with specific forms of behavioural activity linked to defensive reactions or increased attention. In this sense, forward stepping is much more frequently used and is therefore a highly trained locomotor pattern, in contrast to rare backward stepping. In this view, backward locomotion, like other non-stereotyped locomotor patterns, may require specific neuronal control (specific features of rhythm or pattern formation levels of spinal networks or/and specific supraspinal control), and several peculiarities of these two locomotor modes have previously been discovered (Grasso et al., 1998; Harnie et al., 2021). This could be important both for theoretical knowledge about locomotor physiology and for the revision of existing models of locomotor control.

Backward locomotor kinematics is believed to be a simple reversal of forward locomotor kinematics (Thorstensson, 1986; Grasso et al., 1998; Winter et al., 1989; Buford et al., 1990; van Deursen et al., 1998). Currently, the dominant opinion is that forward and backward locomotion share the same neuronal networks (Grillner, 1985; Winter et al., 1989; Buford and Smith, 1990; Duysens et al., 1996; Hoogkamer et al., 2014), and are realised via different combinations of the same CPG elements leading to the different motor patterns. However, despite some comprehensive studies (Buford and Smith, 1990; Buford et al., 1990; Pratt et al., 1996; Rossignol, 1996; Zelenin et al., 2011; Harnie et al., 2021), the mechanisms for backward locomotion have been more poorly investigated than those for forward locomotion, especially in animal models. Since Shik et al. (1966), the decerebrated cat has been widely used for studies of locomotion mechanisms. Our own data on the locomotion of decerebrated cats under epidural stimulation (ES) have suggested that the spinal triggering zone for backward stepping is much narrower (only lumbar segments L6–L7) than that for forward stepping (lumbar segments L3–L7 and sacral segments S1–S2) (Merkulyeva et al., 2018). As proposed previously (Choi and Bastian, 2007; Mahaki et al., 2017), we believe that spinal locomotor networks involved in forward and backward stepping are at least partially separated or can have different supraspinal control. Regardless, the mechanisms for backward locomotion need to be understood. In this study, via electromyography (EMG) and kinematic analysis, we aimed to answer three questions. (1) What is the level of recruitability (ability to be modulated under an external drive)? (2) What is the dependence on the general tonic drive? (3) Is the dependence upon sensory input similar for forward and backward locomotion?

Experiments were carried out on 26 adult cats (Felis catus Linnaeus 1758) of either sex (7 males and 19 females) weighing 2.4–4.5 kg. All animals were reared in a breeding colony on a 12 h:12 h light:dark cycle, with food and freshwater ad libitum, and all procedures were conducted in accordance with a protocol approved by the Animal Care Committee of the Pavlov Institute of Physiology, St Petersburg, Russia, and followed the European Community Council Directive (2010/63EU) and the guidelines of the National Institutes of Health Guide for the Use of Laboratory Animals, Animal Welfare Assurance #A5952-0.

Experimental procedures were described previously (Musienko et al., 2012; Merkulyeva et al., 2018). Cats were anaesthetised deeply with isoflurane (2–4%) delivered in 100% O₂. The level of anaesthesia was monitored by applying pressure to a paw (to detect limb withdrawal), as well as by checking the size and reactivity of the pupils. The trachea was cannulated and the carotid arteries were ligated. The animals were decerebrated at the precollicular-post-mammillary level. All brain tissue rostral to the transection was removed. A laminectomy was performed in the lumbar area for the spinal cord epidural electrical stimulation. The bipolar EMG electrodes (0.2 mm flexible stainless-steel Teflon-insulated wires, AS632, Cooner Wire, Chatsworth, CA, USA) were implanted into hindlimb muscles: m. iliopsoas (IP, one-joint hip flexor, n=21), m. sartorius medialis (SM, two-joint hip and knee flexor, n=7), m. adductor magnus (Add, hip extensor and adductor, n=12), m. biceps femoris posterior (BFP, two-joint hip extensor and knee flexor, n=9), m. rectus femoris (RF, two-joint hip flexor and knee extensor, n=6), m. vastus lateralis (VL, one-joint knee extensor, n=6), m. tibialis anterior (TA, one-joint ankle flexor, n=21), m. gastrocnemius medialis (GM, two-joint ankle extensor and knee flexor, n=17) and m. soleus (Sol, one-joint ankle extensor, n=12).

Anaesthesia was discontinued after the surgical procedures, and the experiments were started 2–3 h thereafter. During the experiment, the rectal temperature, electrocardiography and breathing rate of the animal were continuously monitored.

Decerebrate animals were fixed in a rigid frame; the hindlimbs were positioned on the treadmill, and the distance between the treadmill belt and the fixed pelvis was 21–25 cm (Fig. 1А). Both treadmill belts were moved in a backward or in a forward direction in relation to the cat depending on the locomotor mode tested (forward or backward stepping) with the help of electric motors at a speed of 0.45 m s⁻¹. In 7 cats, one treadmill belt was moving forward and the other backward, eliciting simultaneous forward and backward stepping for different hindlimbs (bidirectional stepping). The stimulation started 2–3 s after the onset of the treadmill belt motion. We showed previously that the optimal location of the ES electrode to induce backward locomotion is in segments L6–L7 (Merkulyeva et al., 2018); therefore, an epidural ball electrode (diameter 0.4 mm) was positioned on the dorsal surface of this spinal region. The following parameters of stimulation were applied: frequency, 5 Hz; pulse duration, 0.2–0.5 ms; current, 100–250 μA (A-M Systems, model 2100). During ES, the forelimbs were restrained. Two mediolateral positions of the ES electrode were used: (i) at the midline (‘central’) and (ii) 1.0–1.5 mm laterally shifted (‘shifted’) (Fig. 1B). We used a shifted position of the ES electrode as in the data about the sophisticated topographic organisation of the spinal cord revealed during intraspinal electrical stimulation (Barthélemy et al., 2007); the authors in particular revealed that sites inducing bilateral locomotor patterns were mainly distributed in more medial sites.

In every cat, anterior–posterior (A–P) limb movements were estimated by the precision single-turn potentiometer sensors, digitised at 5 kHz synchronously with the EMG signals. The angle of rotation of the potentiometer attached to the ankle was proportional to the step length. In all cats, coloured markers were placed on the iliac crest, femoral head, lateral condyle of the femur, knee joint (fixed by Ethilon to the fascia to avoid skin movement), lateral malleolus and fifth metatarsal joint, and the side view of the walking cat (Fig. 1A) was video recorded (50 frames s⁻¹, Basler daA1280-54uc with global shutter, triggered by an external synchronisation signal). For 13 cats, video was processed for an assessment of the angle of the ankle, knee and hip during three phases of the step cycle: maximal load (L), maximal flexion (F) and maximal extension (E). Angle analysis was done only for central ES. To compare the kinematics of locomotor movements, the video recordings were analysed frame by frame (Fig. 1D).

An analysis of ground reaction forces (GRFs) was also performed. Force measurement platforms (350×150 mm) based on single point load cells (Model 1022, 20 kg, Vishay Tedea-Huntleigh, Malvern, PA, USA) were located under the treadmill belts under each limb.

EMG signals were amplified (bandwidth of 10 Hz to 5 kHz, Model 1700, A-M Systems, Sequim, WA, USA) and digitised, as were GRFs and mechanical sensor signals, at 20 kHz with an A/D board (LTR-EU-16, LTR11, L-Card, Moscow, Russia).

We analysed the following kinematic characteristics of the stepping: (1) step period (the total duration of the step cycle); in the case of shifted ES, steps of the hindlimb ipsilateral and contralateral to the ES site were separately assessed; (2) step cycle structure (swing and stance duration); (3) rostro-caudal stability of locomotion of individual hindlimb, estimated using the self-similarity coefficient (the amplitude of the autocorrelation functions at the second peak of the time series of A–P movements of the individual limb; Kim et al., 2007; Fig. 1C); and (4) the asymmetry of stepping periods, estimated as the ratio of the modulo difference between the period of neighbouring steps to their sum. Kinematic features of forward and backward stepping were analysed for all cats – for central ES and for a variable number of cats – for shifted ES (as not all cats were able to do bilateral stepping; see below).

For an analysis of the EMG signal, raw data were digitally filtered in 100–2000 Hz bandwidth. An entire swing and stance duration was divided into 10 bins each, and an integral rectified EMG was calculated in each bin according to the potentiometric signal of the corresponding hindlimb. For one group of cats (n=7), for both central and shifted ES, an integrated EMG activity was used to evaluate the differences between forward and backward stepping. To compare the muscle activity during a particular step phase (swing or stance), for 6–21 cats, its integral EMG value was taken in corresponding bins. For both cases, individual data were normalised to the value of the total EMG activity within the total step cycle. The stability of EMG of individual muscles was estimated using the self-similarity coefficient (see above; EMG smoothed with a 50 ms window was used). The selection of backward and forward trials for the EMG analysis was performed by two experts using the following criteria: the trial should include at least 10 steps, and no pronounced artefacts should be at EMG channels. In every cat, one trial was used for every condition (stepping direction, ES location).

Individual animals were used as an experimental unit (n). Only forward and backward stepping elicited from the same spinal site were analysed, for all conditions (central and shifted ES). All data are presented as means±s.d. As the data were not normally distributed (D'Agostino–Pearson test), a non-parametric criterion was used – Wilcoxon matched-pairs signed rank test. This criterion was used to determine the paired differences (forward versus backward stepping) during the same stimulation condition. A P-value of 0.05 was used as the cutoff for significance. Statistical calculations were performed using Prism 8.0 (GraphPadSoftware, La Jolla, CA, USA).

In general, it is easy to evoke forward stepping, but not backward stepping; therefore, we tested multiple spinal sites to search for an ES point that was successful for evoking both forward and backward stepping (the average number of mapping sites was 6, from 2–14 sites). The optimal spinal regions for evoking backward stepping were located within the locus of the spinal cord dorsal surface corresponding to the L6–L7 segments (as previously determined; Merkulyeva et al., 2018). Five animals were unable to do backward stepping (but had successful forward locomotion); in others, successful bilateral backward stepping could be evoked at least after ES of some dorsal sites. Three cats stepped forward even when the treadmill was moving forward, and in one of them, this stepping was assessed as quite regular (Fig. 2A,B). In this case, the step parameters (step period, step asymmetry, rostro-caudal stability) of this locomotion (termed ‘forward’ locomotion) were different from those for true forward stepping (Fig. 2C–E) (no statistical differences were found because there were few samples).

We used the protocol of the shifted ES (1.0–1.5 mm to the left from the midline) in 11 animals. In contrast to the bilateral forward stepping that was successful in all cats in this study, bilateral backward stepping was elicited in only six animals. For the other five cats, the shifted ES led to backward stepping of the hindlimb ipsilateral to the ES site, not contralateral. Therefore, for most cases, only data for the ipsilateral hindlimb are presented.

In one cat with unilateral backward stepping, the shifted ES led to alternating periods of unilateral stepping: the left hindlimb made 4–8 steps solely when the right hindlimb was not moving. Then, the right hindlimb started moving while the left was not moving (Fig. 2G). At the same time, regular bilateral forward stepping was evoked from the same ES site (Fig. 2H). The step durations of the two hindlimbs during backward steps were different but equal, in the case of forward stepping (Fig. 2F) (no statistical differences were found because there were few samples).

In four out of 11 cats, the ES electrode was moved not only to the left (Fig. 3B) but also to the right (Fig. 3C) from the midline. In all cats, under these conditions, bilateral forward stepping was elicited, leading to reciprocal movements of the hindlimbs and a clear reciprocity in the activity of the left and right TA muscles (Fig. 3). For the backward stepping, normal interlimb reciprocity could be obtained only during central ES (Fig. 3A); during shifted ES (irrespective of the side of the shift), this reciprocity was impaired (Fig. 3B,C).

One of the differences found between forward and backward stepping was the ability to initiate rhythmical activity after ES was switched on (activity delay) and to continue it after ES was switched off when only the treadmill belts were moving (after-steps). A representative example of these features for forward and backward stepping evoked by the same ES site is shown in Fig. 4A. Paired data for forward and backward stepping evoked from the same ES sites were analysed, and the data as a percentage of forward stepping are shown at Fig. 4B,C. Steps were analysed based on the appearance of a locomotor pattern on the potentiometers and burst activity on the EMG channels. The activity delay was assessed as a gap between the first current pulse and the first step (just the step but not artefact dragging of the hindlimb when the ES was switched off – this bilateral artefact movement is clearly visible in Fig. 4A for forward stepping, marked by black triangles at potentiometric curves).

During central ES, the first steps for forward stepping began 0.2–3.0 s after ES was switched on, and for backward stepping, it began at 0.9–6.0 s. During shifted ES, the first steps for forward and backward stepping began at 0.2–2.8 s and 0.6–5.4 s, respectively. Only paired data (forward and backward episodes evoked from the same ES point) were used for statistical comparison. In 16/19 cats for central ES (P<0.0001; Wilcoxon test) and in 9/10 cats for shifted ES (P=0.0371; Wilcoxon test), the first forward steps were detected earlier than the backward steps (Fig. 4B). For both ES types, these differences were significant. As mentioned above, only data for the ipsilateral hindlimb are presented.

During central ES, the number of after-steps ranged from 0 to 21 steps for forward stepping and from 0 to 3 steps for backward stepping. For the shifted ES, it ranged from 1 to 11 steps and from 0 to 2 steps, respectively, for forward and backward stepping. All animals (20/20 and 11/11, respectively) had more forward after-steps compared with backward after-steps for both ES types (Fig. 4C); these differences were significant (P<0.0001 and P=0.0010; Wilcoxon test). We also found different forms of the potentiometric signal after ES was switched off: for forward stepping, a gradual decrease in the amplitude of steps was visible (Fig. 4A, upper right); for backward stepping, a sharp termination was obtained (Fig. 4A, lower right). Thus, clearly, rhythmical activity began after stimulus onset and ceased more quickly after removal of the stimulus for backward stepping.

During central ES, in 18/20 cats, the period of backward stepping was longer than that for forward stepping for both hindlimbs (P<0.0001 for left hindlimb, P<0.0001 for right hindlimb; Wilcoxon test) (Fig. 5A). In all cats, except for one cat showing different effects in two legs, the swing phase for backward stepping was significantly longer than that for forward stepping, for both hindlimbs (P<0.0001 for left hindlimb, P<0.0001 for right hindlimb; Wilcoxon test) (Fig. 5C). In 13/20 cats, the stance phase for backward stepping was longer than that for forward stepping, but it was significant only for the left hindlimb (P=0.0296 for left hindlimb, P=0.1429 for right hindlimb; Wilcoxon test) (Fig. 5B). During shifted ES, the period of backward stepping was longer than that of forward stepping for the hindlimb ipsilateral to ES in 8/10 cats as well as for the hindlimb contralateral to ES in 7/10 cats, but there were insignificant differences (P=0.0645 and P=0.1484; Wilcoxon test) (Fig. 5A). In 9/10 cats, the swing phase for backward stepping was longer than that for forward stepping, for the ipsilateral hindlimb (P=0.0137; Wilcoxon test), and in all cats for the contralateral hindlimb (P=0.0078; Wilcoxon test) (Fig. 5C). The stance phase was randomly longer or shorter during backward stepping (P=0.4922 for the ipsilateral hindlimb, P=0.5469 for the contralateral hindlimb; Wilcoxon test) (Fig. 5B). A more clear difference between the duration of the step period of the two hindlimbs (period delta) was obtained for backward stepping (P=0.0056; Wilcoxon test).

To reveal the rostro-caudal stability of hindlimb movements, the coefficient of self-similarity was used. During central ES, for both hindlimbs, in 18/20 cats, this coefficient was lower for backward stepping (P=0.0024 for the left hindlimb, P=0.0005 for the right hindlimb; Wilcoxon test) (Fig. 5D). In our previous study (Merkulyeva et al., 2018), using a smaller number of animals (n=6), we were unsuccessful at revealing these differences. For the shifted ES, stability was lower during backward stepping for both hindlimbs in all cats studied (P=0.0039 for the hindlimb ipsilateral to the ES, P=0.0078 for the hindlimb contralateral to the ES; Wilcoxon test) (Fig. 5D).

Hip, knee and ankle angles were measured during three phases of the step cycle: maximal load, maximal flexion and maximal extension. The hip angle during the load and extension phases but not the flexion phase was significantly smaller during backward stepping (P=0.0002, P=0.0002 and P=0.588, respectively; Wilcoxon test) (Fig. 5E). The knee angle was significantly bigger for the load phase but not for the extension and flexion phases during backward stepping (P=0.0002, P=0.187 and P=0.99, respectively; Wilcoxon test) (Fig. 5F). The ankle angle was significantly smaller for the extension phase but not for the load and flexion phases during backward stepping (P=0.0002, P=0.244 and P=0.589, respectively; Wilcoxon test) (Fig. 5G).

To compare muscle activity within the step cycle, averaged percentage EMG curves (as a percentage of the overall EMG activity during the total step cycle) were analysed for forward and backward stepping during central and shifted ES for 7 cats. Only three muscles: the IP and TA flexors, and the GM extensor were compared because they were implanted in both the left and right hindlimbs. A clear reciprocity in the EMG activity of the hindlimb flexors was obtained for both ES for forward stepping (Fig. 6A,C,E,G) and only for the central ES for backward stepping (Fig. 6B,F). Instead, despite different phase characteristics during backward stepping (less preference for the swing–stance boundary compared with forward stepping), the GM extensor muscle showed clear reciprocity in the activity of two hindlimbs irrespective of ES type (Fig. 6I–L).

In our previous paper, particular differences in the total activity of several muscles – the IP, Add, RF and VL – were obtained for forward and backward stepping (Merkulyeva et al., 2021). In this study, we compared the integral EMG activity of the IP, SM, BFP, TA, GM and Sol. If both muscles (left and right) were implanted (for IP, TA and GM), both values are presented. As previously, in all cats (but in three cats, in one leg only), the IP was less active during backward stepping (P<0.0001; Wilcoxon test) (Fig. 7A). The EMG signal of the SM was randomly either higher or lower during backward stepping (P=0.375; Wilcoxon test) (Fig. 7A). The BFP was more active during backward stepping in 8/9 cats (P=0.0117; Wilcoxon test) (Fig. 7A). In 19/21 cats, the TA signal was lower during backward stepping (P<0.0001; Wilcoxon test) (Fig. 7A). In 15/17 cats, the GM was more active during backward stepping (P<0.0001; Wilcoxon test) (Fig. 7A). Sol activity was either higher or lower during backward compared with forward stepping (Fig. 7A). A summary of muscle activity is presented in Table 1, where six main functions of hindlimb muscles are illustrated: hip flexion (IP, SM and RF), hip extension (Add), knee flexion (SM, BFP and GM), knee extension (RF and VL), ankle flexion (TA) and ankle extension (GM and Sol).

Because there was an increase in the hip and ankle extensor activity during backward stepping, we also compared the GRFs during forward and backward stepping: no significant difference was obtained (P=0.362; Wilcoxon test).

For numerous muscle samples, muscle stability was assessed. In most cats, stability of the IP and TA was lower for backward stepping than for forward stepping (IP: 15/19 cats, P=0.0024; TA: 13/19 cats, P=0.0361; Wilcoxon test); no statistically significant differences were obtained for the GM, Sol, BFP and SM (GM: P=0.154; Sol: P=0.064; BFP: P=0.1484; SM: P=0.3125; Wilcoxon test) (Fig. 7B).

The sensory input from limbs is crucial for stable stepping (McGibbon et al., 2005; Musienko et al., 2007). We mismatched the sensory inputs to the left and right hindlimbs in seven cats, evoking bidirectional stepping when one treadmill belt was moving forward and another was moving backward, eliciting simultaneous forward and backward stepping for different hindlimbs. Two variants of bidirectional stepping were used: with central and shifted ES (n=7). After-steps were assessed only for central ES. These data were in line with data obtained for unidirectional forward and backward stepping: hindlimb stepping forward resulted in more after-steps compared with hindlimb stepping backward in all cats (P=0.0157; Wilcoxon test) (Fig. 8A,B). In most cats stepping bidirectionally (6 of 7 cats), the number of after-steps of hindlimbs stepping forward was shifted towards the values for the backward stepping (hindlimb stepping forward during bidirectional stepping made 24–83% after-steps compared with unidirectional forward). For hindlimb stepping backward during bidirectional stepping, no shift was obtained.

At the same time, the main feature of backward stepping – less integrated EMG activity of the flexor muscles and higher activity of the extensor muscle (see Fig. 7) – was observed only for the hindlimb stepping backward during bidirectional locomotion [it was statistically significant for TA and GM (P=0.0313 and P=0.0313; Wilcoxon test) but not for IP (P=0.4688; Wilcoxon test)].

In all cats tested, shifted ES led to either backward stepping only in the hindlimb ipsilateral to the ES (see example in Fig. 8D) or asymmetric backward stepping with the predominance of the hindlimb ipsilateral to the ES. Moreover, successful bilateral forward stepping was obtained (Fig. 8C). For bidirectional stepping, two variants could be evoked: when a backward moving hindlimb was located ipsilateral or contralateral to the shifted ES site. An interesting phenomenon was revealed: successful bidirectional stepping could be evoked only in the first case (Fig. 8E). In the opposite case, mainly unilateral forward stepping could be elicited with no rhythmic (or with non-periodic) movements of the hindlimb contralateral to the ES site (Fig. 8F). Moreover, the integrated EMG activity was lower for the contralateral flexors (in all cats for IP, and in 6/7 cats for TA) irrespective of the stepping direction (P=0.0156 and P=0.0156 for IP; P=0.0156 and P=0.0313 for TA, for forward and backward stepping, respectively; Wilcoxon test) (Fig. 8G,H). For the contralateral extensor (GM), it was more, less or equally active compared with the ipsilateral GM, irrespective of the stepping direction (P=0.5781 and P=0.3750, for forward and backward stepping, respectively; Wilcoxon test) (Fig. 8I).

In this study, we aimed to evaluate the peculiarities of backward locomotion regarding rhythmicity, bilateral interconnections, influence of the tonic drive and sensory input. Only paired ‘backward stepping–forward stepping’ data were analysed.

Previously, we showed that the spinal triggering zone for ES-evoked backward stepping in cats is much narrower than that for forward stepping (caudal part of L5–L7 segments versus the overall lumbosacral region) (Merkulyeva et al., 2018). However, even when using the ES of the ‘correct’ segments, it was more difficult to initiate coordinated backward stepping compared with forward stepping in acute decerebrate cats, and several cats were unable to perform backward stepping. If induced, backward stepping had lower rostro-caudal stability and lower stability of EMG signals compared with forward stepping. Less regularity of the step cycle during backward stepping was obtained previously in cats spinalised at thoracal segment T13 (Barbeau et al., 1987). Additionally, backward stepping was characterised by fewer or no after-steps compared with forward stepping, and had a longer activity delay. The general duration of the activity delay during forward stepping corresponded to that previously observed during ES of decerebrated cats (Iwahara et al., 1992). These data possibly indicate that neuronal networks responsible for backward stepping are less easily recruited and less inclined to self-sustaining rhythmic activity than those responsible for forward stepping. Backward locomotion has been documented in animals after complete spinal cord transection at T12–T13 (cats) or T7 (rats) (Barbeau et al., 1987; Courtine et al., 2009; Harnie et al., 2021), suggesting that this locomotor mode is possible without descending inputs from the brain. Barbeau et al. (1987) and Harnie et al. (2021) also documented that additional stimulation (perineal stimulation) is needed to evoke backward but not forward stepping. However, in natural life, despite a high degree of autonomy, CPGs are under the influence of supraspinal centres initiating and modulating their activity to behavioural tasks or environments (Shik and Orlovsky, 1976; Grillner, 1985; Buford and Smith, 1990). We showed no unambiguous evidence for backward stepping under the control of a specific CPG that was different from that for forward stepping; however, some facts such as no after-steps for backward stepping, pronounced difficulty in backward locomotion initiation and a much narrower spinal triggering zone for backward stepping compared with forward stepping indicate the strong peculiarities in the networks responsible for the control of backward stepping [those responsible for rhythm generation, those responsible for pattern formation (according to McCrea and Rybak, 2008), or both]. Another possibility is that backward stepping is dependent in a different way upon the supraspinal commands compared with forward stepping and, in particular, supraspinal commands can inhibit the ability to perform backward stepping.

We also found that the total duration of the step cycle for backward stepping was longer than for forward stepping, and a more obvious difference in the step duration of backward and forward stepping was observed for the swing phase. This indirectly suggests that the neuronal networks responsible for rhythm generation work slower during backward stepping. Interestingly, previously, an opposite result was found: a shorter locomotor cycle was found for backward stepping (Barbeau et al., 1987; Buford and Smith, 1990), but intact or spinal cats were used in those studies. These data again suggest the high peculiarity of the locomotor network activity in decerebrate animals: under this condition, locomotor networks no longer have full supraspinal control as in intact animals but have not yet totally lost supraspinal input as in spinal animals. Regarding the peculiar behavioural role of the backward stepping, we propose that there is a deficit in the supraspinal tonic drive in decerebrate animals, reducing the ability of this locomotor mode.

Why do we pay attention to the supraspinal input if backward locomotion is possible in spinal animals? Upon investigating backward mechanisms in spinal cats, it has been proposed that the hip position determined by the treadmill direction (extension during backward treadmill belt movement and flexion during forward treadmill belt movement) is essential for the initiation of forward or backward stepping, respectively (Harnie et al., 2021). However, in three of our cats, regardless of treadmill direction, only forward locomotion was elicited. Longitude locomotion is based upon the prolonged activity of the neurons that are able to generate a basic rhythmic motor output under the excitatory supraspinal drive that presumably originates from brainstem reticulospinal neurons, particularly from the so-called mesencephalic locomotor region (MLR) (Garcia-Rill and Skinner, 1987; Noga et al., 2003; Matsuyama et al., 2004). Electrical stimulation of the MLR leads to forward but not backward locomotion. Possibly, evocation of locomotion in the incorrect direction is a result of a disbalance in sensory input and the descending input from the brain structures in decerebrated animals. Perhaps the less the MLR influences the spinal locomotor networks, the higher the ability to use only the sensory input from the treadmill.

Multiple supraspinal inputs contribute to spinal CPGs; of special interest regarding backward locomotion is the periaqueductal grey (PAG) structure responsible for motor behaviour and defensive reactions (in particular, backward defensive behaviour) characterised by upright postures and backward movements (Bandler and Depaulis, 1991; Depaulis et al., 1992). Interestingly, in cats, the PAG has a direct projection to the spinal cord and the lumbar region in particular; the densest descending fibres were observed within laminae VII and VIII (Mouton and Holstege, 1994; Vanderhorst et al., 1996) where more c-fos-stained nuclei were obtained after backward stepping than after forward stepping (Merkulyeva et al., 2018). It is possible that spinal networks per se can generate locomotion in any direction under the influence of sensory input from the hindlimb's skin, muscles and joints. However, the more complicated the motor behaviour, the more complicated the supraspinal control (in particular, those depending on other sensory modalities) contributing to several different functions, such as active escape (quick galloping) and backward defensive behaviour, when needed. Interestingly, highly specialised creatures, such as blind naked mole rats, have lost their vision and hearing and mainly rely upon tactile sense but are equally able to perform forward and backward movements (Eilam and Shefer, 1992). In general, backward stepping seems to be more flexible and less automated compared with forward stepping.

Another peculiarity of backward stepping is lower inter-limb coordination compared with forward stepping. Successful stepping is based on the interconnections between the two CPGs located within the opposite halves of the spinal cord (each controlling the ipsilateral hindlimb; Kato, 1990). The basis for this interconnection is commissural interneurons that have axons crossing the midline of the spinal cord (Marder and Calabrese, 1996; Kiehn and Butt, 2003). Even in the case of the central ES, some differences between forward and backward stepping were observed; in particular, the higher delta between the duration of the step period of the two hindlimbs. In the shifted ES, a reduction in the inter-limb reciprocity for hindlimb muscles was obtained for backward but not forward stepping, coupled with general impairment of backward stepping bilaterality. Moreover, bidirectional locomotion in decerebrated cats (Lyakhovetskii et al., 2021) can be elicited during the shifted ES when the treadmill belt moves forward (eliciting backward stepping) and is located at the side ipsilateral to the ES. The leading role in the mechanisms of ES-evoked stepping has been assigned to the direct activation of myelinated proprioceptive dorsal root fibres polysynaptically projecting to spinal locomotor networks (Struijk et al., 1993; Gerasimenko et al., 2003; Capogrosso et al., 2013). However, unlike the dominant opinion, a direct effect of the ES on the tracts and spinal interneurons has also been proposed for cats and rats (Lavrov et al., 2006; Edgerton and Harkema, 2011). Regardless, ES electrode displacement shifts the balance between either sensory inputs or neuronal networks located within the lateral and medial areas of the grey matter. Medial areas of the grey matter contain commissural neurons contributing to left–right hindlimb alternation during locomotion (Matsuyama et al., 2004; Kiehn and Butt, 2003). Previously, Barthélemy et al. (2007) induced bilateral flexion and bilateral extension mainly by intraspinal stimulation of cat medial grey matter sites, not lateral ones. Thus, we can expect a particular disruption in the bilateral locomotion during ES displacement; however, it was observed only during backward but not forward stepping, which allows us to propose some peculiarities of the neuronal networks controlling backward stepping.

In intact cats, the swing in backward locomotion is accompanied by hip extension and knee flexion in contrast to hip flexion and knee flexion observed during forward stepping (Perell et al., 1993). This corresponds well to the reduced activity of hip flexors IP and RF and the increased activity of hip extensors Add and BFP during backward stepping (note that the BFP muscle is mainly a knee flexor, not a hip extensor; English and Weeks, 1987; Chanaud and Macpherson, 1991). Additionally, a contributor to horizontal displacement of the body during backward locomotion is knee extension, in contrast to hip extension during forward locomotion (Buford et al., 1990). Hip angles are significantly less, and knee angles are significantly higher during the load stage for backward stepping, and both angles are unchanged during maximal flexion phase. More knee flexion was found in intact cats during backward locomotion (Buford et al., 1990), and higher EMG activity was obtained for the BFP and GM knee flexors. The SM is simultaneously the hip and knee flexor (Hoffer et al., 1987), and it can show both increased and decreased EMG signals during backward stepping, possibly because of the differences in the two functions in individual cats. During backward locomotion, the ankle joint was found to be less flexed during extension compared with that during forward locomotion in intact cats (Buford et al., 1990), and in this study, the EMG signal of the ankle flexor TA was lower and that of the ankle extensors GM and Sol was greater during backward stepping. All these data suggest that the kinematic features of backward stepping in decerebrated cats are similar to those in intact cats.

Previously it was hypothesised that the same network is responsible for forward and backward stepping, ‘…by just reversing the hip–knee coordination by a phase shift of 0.5 (hip flexion concomitant with knee and lower leg extensor activity) …’ (Grillner, 2011). Several studies on intact humans (Duysens et al., 1996; Thorstensson, 1986; Winter et al., 1989) and intact cats (Buford and Smith, 1990) have led to the supposition that backward locomotion is only a reversal of forward locomotion. However, one study showed no transfer of forward training to untrained (backward) locomotion in spinal cord injured patients (Grasso et al., 2004), another showed that walking adaptations are independent for leg walking forward and backward in intact humans (Choi and Bastian, 2007), and another described the different EMG activation profiles for the two locomotor modes in intact humans (Lay et al., 2007; Mahaki et al., 2017). In our previous studies on decerebrated cats, we found that the regions responsible for initiating backward and forward stepping are distributed differently over the lumbosacral area (Merkulyeva et al., 2018, 2021). Additionally, there are current data on the differences in rhythmogenic abilities, bilateral interconnections, etc., that suggest several peculiarities in spinal circuitries controlling these two locomotor modes. Several possibilities are proposed: (1) backward locomotor circuits can be spread over the lumbosacral region that are locally dense only within the L6–L7 segments; (2) backward locomotor circuits can be inhibited by reticulospinal influence in favour of the main and more developed locomotor pattern: forward stepping; and (3) backward locomotor circuits can have direct or indirect specific supraspinal input, possibly from brain structures related to the limbic system. Further studies are needed to understand the neural control of locomotor activity in different directions, which are of fundamental interest and may be clinically relevant because multi-direction stepping training is a useful paradigm employed for the treatment of a variety of orthopaedic and neurological diseases (Błażkiewicz, 2013; Shah et al., 2012; Shapkova et al., 2020).

The authors thank Polina Shkorbatova for help with animal care.