The control of posture and locomotion requires the sequential and, at least in the case of locomotion, rhythmic activation of series of skeletal muscles bilaterally distributed along the body. Recruitment of these muscles during locomotion is primarily ensured by basic motor commands generated by the so-called central pattern generators (CPGs; Figure 1A; for recent reviews, see Côté et al., 2018; Grillner and Kozlov, 2021). Briefly, basic commands are organized by sets of excitatory interneurons (INs), which are arranged in modules generating fundamental recurring activity, and which are coordinated through inhibitions implicating a variety of local inhibitory INs (Cangiano and Grillner, 2005; Grillner et al., 2005; Berkowitz et al., 2010; Bagnall and McLean, 2014; Jay and McLean, 2021). Bilateral coordination depends specifically on commissural INs while ipsilateral (e.g., flexor/extensor) coordination depends on ipsilateral INs. These interneuronal networks connect motor neurons (MNs) that transmit central motor commands to effector muscles. However, MNs are not just passive output neurons. As it is the case for INs in the CPG, MNs possess some particular membrane properties, such as the persistent sodium current (I_NaP; Tazerart et al., 2007, 2008), allowing them to integrate rather than just follow the numerous synaptic inputs they receive. Locomotor CPGs are localized in the spinal cord, and their organization depends on the animal biomechanical apparatus.

In fish or larval amphibian, both propulsion and posture are ensured by axial muscles that are organized in myotomes distributed along the body axis (D’Elia and Dasen, 2018) and controlled by bilaterally alternating neuronal networks, segmentally iterated throughout the spinal cord (Roberts et al., 2010; Berg et al., 2018). In contrast, in quadrupeds the CPGs dedicated to fore- and hindlimb movements are respectively segregated in cervical and lumbar segments (e.g., in rats: Ballion et al., 2001, and Cazalets et al., 1995, respectively), coordinated through propriospinal pathways (Juvin et al., 2005, 2012), and are putatively organized in interconnected modules, each responsible for the command of flexion/extension cycles at a single joint (Grillner and El Manira, 2020; Grillner and Kozlov, 2021). However, the exact nature and organization of CPGs in mammals remain unclear, although the rise of genetic approaches has already allowed the identification of several populations of INs putatively involved in rhythm-generating GPGs (Goulding, 2009; Rancic and Gosgnach, 2021).

The postural nervous system, in contrast, is largely distributed along the spinal cord. Although variations exist between species and in accordance with the type of posturo-locomotor movements they produce, each spinal segment participates in organizing the axial muscle contractions responsible for postural activities (reviewed in Guillaud et al., 2020). Globally, postural control can be separated into two major functions: one is dedicated to body stabilization in the absence of goal-directed movement and resists external perturbations including gravity, while the other is dynamically engaged during self-motion and ensures body balance and orientation in space. Usually, the control of posture, whether it concerns perturbation-induced reflexes or gait-related control, is considered to depend mainly on descending commands from supraspinal centers (reviewed in Takakusaki, 2017), partly because the available knowledge about the organization and dynamic activity of spinal postural networks is scarce. In fact, most data come from analyses of thoracic back muscles during locomotion. In rats (Falgairolle and Cazalets, 2007) as in humans (de Sèze et al., 2008), metachronal propagations of back muscle contraction were observed to occur with comparable patterns during walking. Moreover, such metachronal waves were found to persist in thoracic ventral root recordings during pharmacologically induced fictive locomotion in the in vitro isolated spinal cord of newborn rats, suggesting the existence of a propriospinal coordination between the CPG for locomotion and the thoracic networks (Falgairolle and Cazalets, 2007). However, similarly to fishes it remains quite impossible to ascertain that such back muscle activities are not also simply related to propulsion, which is made possible by the vertebral spine flexibility observed in the majority of vertebrate species. For example, during cheetah or greyhound run, cycles of spine curvature/extension strongly participate in the animal’s propulsion (see Hudson et al., 2012).

Nonetheless, the particular anatomy of the vertebral column of post-metamorphic anuran Xenopus laevis (Beyeler et al., 2008) strongly limits spine flexibility. Indeed, propulsion is entirely provided by the hindlimbs, whereas the contraction of the axial back muscles dorsalis trunci generates only small twists of one vertebra with respect to the neighboring others, conferring these muscles a purely postural function (at least for myomeres 3 and 4 since their insertion on the skeleton prohibits any participation in propulsion; Olechowski-Bessaguet et al., 2020). Electrophysiological recordings from either back muscles in vivo or dorsalis-innervating thoracic ventral roots in vitro showed strict coordination with hindlimb muscles/lumbar ventral roots activity during swimming, and thoracic MNs were likely directly activated by the lumbar CPG (Beyeler et al., 2008). Thus, in juvenile Xenopus at least, axial postural networks are subdued to propulsive networks through segmental propriospinal connections when the locomotor CPG is active, in order to generate the appropriate anticipatory postural adjustments necessary to ensure accurate locomotion.

Although intrinsic auto-organization of posturo-locomotor activities exists within spinal networks, these are modulated by both local sensory feedbacks and descending commands from supra-spinal structures (Figure 1; Grillner and El Manira, 2020). During locomotion, local sensory control is mediated by segmental proprioceptive and cutaneous afferents in a cycle-dependent manner (MacKinnon, 2018). For instance, load-mediating afferents (Ib) are implicated in joint extension while afferents that inform about muscle length variation (Ia afferents from muscle spindles) are mainly involved in joint movement termination and initiation of movement in the opposite direction, acting on the reciprocal inhibition between flexor and extensor (e.g., in humans: Perez et al., 2003). Cutaneous information is involved equally in locomotor and postural activities; notably during locomotion, they modulate the cycle by interacting presynaptically with other afferent signals. In turn, during locomotion sensory feedback afferents are cyclically modulated by the CPG via specialized inhibitory INs (reviewed in Côté et al., 2018). Descending commands from supra-spinal centers are also able to modulate spinal reflexes in a phase-dependent manner. In humans for example, increasing the postural threat during walking changes the gain of the H-reflex (Krauss and Misiaszek, 2007), which suggests that convergence of both local sensory and descending brain commands onto spinal INs (Stecina and Jankowska, 2007; Stecina et al., 2008) is essential to regulate local reflexes (Phadke et al., 2009). Nevertheless, even in spinalized cats (deprived of every supra-spinal inputs) walking on a treadmill, pace can gain symmetry if treadmill velocity is increased (Dambreville et al., 2015), indicating that spinal networks are nevertheless intrinsically able to adapt the motor program they produce as gait increases in speed, and this probably via the implication of the local, symmetrically organized sensory feedback networks.

Supra-spinal structures determine initiation and termination of locomotion, as well as gait adaptation to challenging milieus, in a cascade of commands sequentially involving cortical and thalamic structures, basal ganglia, mesencephalic nuclei and, finally, reticulospinal nuclei that directly control spinal CPGs (Grillner et al., 2008; Kozlov et al., 2009; Leiras et al., 2022). They are also responsible for a majority of postural commands, besides the propriospinal coordinating pathways involved in locomotion-required postural adjustments (see above). Among these structures, sensory-motor cortices, via bilateral crossed corticospinal tracts, are mostly implicated in the voluntary control of fine body and limb motion (see Takakusaki, 2017). In animals in which the cortex is much less developed, the fine control of movement seems to be achieved by the red nucleus (Olivares-Moreno et al., 2021), which is intimately interconnected with both the cortex and the cerebellum (Cacciola et al., 2019). Basal ganglia also participate in posture regulation, through projections on brainstem motor centers, including the mesencephalic locomotor region (MLR) and the reticular formation (Takakusaki et al., 2016). Within the brainstem (Figure 1B), vestibulospinal (VS) and reticulospinal (RS) nuclei play the major role, reticulospinal neurons relaying the great majority of higher posturo-locomotor commands from the MLR (as described in many details in the lamprey; see Le Ray et al., 2011) and cortex (Stecina and Jankowska, 2007) as well as inputs from the central vestibular nuclei (CVNs; e.g., Deliagina et al., 1992; Pflieger and Dubuc, 2004). Yet, in all vertebrates vestibulospinal neurons also directly activate spinal motor networks to ensure body postural adjustments and maintain head position in space (Straka and Baker, 2013; Cullen, 2016; Bagnall and Schoppik, 2018). In addition, CVN neurons affect indirectly the control of posture via their projections to the thalamus and insular cortex, as well as via their bilateral interactions with cerebellar nuclei (see Takakusaki, 2017; Cullen, 2019). The latter also directly and indirectly integrate vestibular afferents (Carpenter et al., 1972; Schniepp et al., 2017).

Many studies investigated the organization of vestibulospinal projections in various species, from lamprey (Deliagina et al., 2014) to mammals (Kasumacic et al., 2015); however, only a recent work in the juvenile Xenopus clearly provided a functional and anatomical description of vestibulospinal projections onto spinal neurons specifically involved in the control of posture (Olechowski-Bessaguet et al., 2020). This study reports that vestibulospinal nuclei activate identified postural MNs through two distinct ways: the first, classically described in all species, consists of a direct activation of populations of spinal neurons (INs and MNs) involved in the control of posture at the segmental level (e.g., thoracic MNs activating dorsalis muscles), and conveys principally signals related to head position; the second, yet undescribed, consists of an indirect pathway to activate these same thoracic neurons via a lumbar interneuronal relay dispatching the vestibular command to thoracic and lumbar MNs simultaneously, and conveys mainly velocity signals from the vestibular system.

Whatever the species, central vestibular neurons integrate spatially corresponding, convergent vestibular, optokinetic, proprioceptive and cutaneous inputs, and it has been postulated that the vestibulospinal network anatomical organization included all sensory-motor transformations appropriate for the control of posture (Vidal et al., 1993). Interestingly, although working according to a “push-pull” mechanism (one side being inhibited when the other side is excited by a given direction of head movement) the vestibular system is symmetrically organized, and mathematical models have suggested that the bilateral symmetry of vestibular reflexes (notably the vestibulo-ocular reflex; Smith and Galiana, 1991) resulted from this symmetry in the vestibulo-motor system (e.g., Ris et al., 1995). In particular, vestibulospinal projections are anatomically organized in symmetry groups likely related to their physiological sensory-motor function (McCollum, 2007), providing the vestibulospinal system with a powerful capacity to adapt precisely spinal posturo-locomotor functions. However, similarly to segmental sensory feedbacks in the spinal cord, vestibular integration is subject to modulation during active movement (Chagnaud et al., 2015), notably via interactions with the cerebellum (Takahashi and Shinoda, 2021) and via bilateral ascending spino-bulbar pathways conveying an efference copy of the motor program generated by spinal networks. These latter ascending efferent copies were clearly demonstrated in both larval (Combes et al., 2008) and juvenile (von Uckermann et al., 2013) Xenopus to coordinate directly eye movements with locomotor-induced head displacements, and were proposed to exert side-specific filtering of vestibular sensory integration in the CVNs (Straka et al., 2018). Recent studies in human subjects provided evidence that comparable efference copy mechanisms may also account for the stabilization of eye vertical position, at least during fast locomotion (Dietrich and Wuehr, 2019; Dietrich et al., 2020).

It has been long known that damages in the human sensory-motor cortex produce side-specific effects (Robinson, 1979), with acute lesions on the right side affecting mostly the control of posture while left side brain injuries rather generate apraxia (Spinazzola et al., 2003). As a primary response to unilateral stroke in the motor cortex, deep cortical reorganization occurs, as recently reported in patients with focal ischemic stroke where theta burst-induced depression was initially higher in the contralesional motor hemisphere and slowly recovered during the following weeks, whereas no changes were observed in the ipsilesional hemisphere (Hordacre et al., 2021). In rodents, stroke on one side of the sensory-motor cortex similarly triggers contralateral motor cortex plasticity (Takatsuru et al., 2009), but may as well affect the ipsilesional cortex. As shown in the adult rat, focal ischemic stroke in the motor cortex area controlling a forelimb initially generates motor deficits that are progressively compensated for, due to ipsilesional motor cortex neurons involved in hindlimb control sprouting new collaterals into cervical spinal motor circuitry (Figure 2; Starkey et al., 2012).

In contrast, a developmental alteration of the cerebellum symmetrical organization (unilateral hypoplasia) does not significantly modify postural and locomotor movement symmetry (Immisch et al., 2003) but globally slows down movements and delays motor skill acquisition in children (Benbir et al., 2011), suggesting that cerebellar symmetry is not a pre-requisite for accurate motor control. However, it has been shown in primates that, depending on the considered cerebellar network, a unilateral acute lesion oppositely affects nystagmus direction (Romano et al., 2020), suggesting that posture and eye movement rely on distinct neural organizations. In rats, posture is comparably affected by acute lesions of the sensory-motor cortex on either side; however, the reported asymmetry in hindlimb postural response seems to rely on distinct mechanisms, a right-side damage triggering spinal motor network reorganization, and a left-side lesion instead altering local sensory integration (Zhang et al., 2020).

As a result of spinal cord injury (SCI), disconnection of descending pathways that carry supra-spinal motor commands toward spinal motor networks generates postural and locomotor disturbances (e.g., reduced use of ipsilesional limb after unilateral rubrospinal tract lesion; Webb and Muir, 2003) that can be (partially) compensated for by downstream (see later) and/or upstream plasticity. Such plasticity can occur spontaneously but usually benefits from sensory-motor training similarly in patients and animals (Knikou, 2010; Torres-Espín et al., 2018; but see Fouad et al., 2000). In rats, SCI does not trigger damaged corticospinal neurons to degenerate (Nielson et al., 2010) but these are likely redeployed in new functional cortical networks. For instance, incomplete SCI at a cervical level leads to a transient loss of ipsilateral paw representation in the primary somatosensory cortex; this representation is reactivated (Ghosh et al., 2009; Martinez et al., 2010; Bazley et al., 2014) and tactile ability ameliorates with sensory-motor training protocols, mobilizing the preserved pathways and resulting in primary motor cortex reorganization (Martinez et al., 2009, 2010). On the contralesional side, consecutive to cord hemisection and bilateral dorsal column cut, motor cortex spiny pyramidal neurons exhibit tapered and longer dendrites, due to an overexpression of polysialylate cell adhesion molecules that limit synapse formation (Kim et al., 2006, 2008), indicating cellular reorganization in cortical networks. Finally, plasticity may occur as well in the lower brainstem, as recently investigated in lampreys (Hough et al., 2021) where spinal cord hemisection leads ipsilesional RS neurons to adapt their electrical properties, compared to contralateral uninjured RS neurons. However, such changes in intrinsic properties do not seem to affect the way these command neurons integrate input signals, notably sensory trigeminal inputs, raising the question of the functionality of such a plasticity and whether this participates in subsequent locomotor recovery.

Spinal cord injured animals are common models for studying neural plasticity (Dietz and Fouad, 2014; Alizadeh et al., 2019) in the context of motor functions (respiration: Streeter et al., 2020; posture and locomotion: Brown and Martinez, 2019). Most species are capable of partial or complete recovery after SCI, demonstrating the intrinsic ability of spinal sensory-motor networks to organize functionally relevant motor commands even in the absence of part or all descending commands and neuromodulation (Rossignol et al., 1996; but see also Merlet et al., 2021). Lumbar SCIs are more deleterious than mid-thoracic ones as reported in cats (Rossignol et al., 2002) and rats (García-Alías et al., 2006), probably because they may directly injure the locomotor CPG. In this context, spinal cord hemisection or incomplete SCI at a cervical or thoracic level provide interesting models to investigate the mechanisms by which a bilateral symmetrically built CNS adapts to abnormal imbalance, in order to maintain symmetrical biomechanics as accurate as possible. For instance, whereas a cervical cord hemisection induces the operational loss of the ipsilesional forelimb in rats, both hindlimbs recover coordinated activity compatible with functional locomotion (Ghosh et al., 2009).

In cats with unilateral thoracic SCI, after intensive motor training both the spinal CPG and sensory-motor connections exhibit deep remodeling (Barrière et al., 2008; Martinez et al., 2013), which results in the generation of motor command asymmetries downstream of the lesion to restore bilateral locomotion. In addition, such reorganized spinal networks persisted after a subsequent complete spinal cord transection, the latter transiently triggering opposite asymmetries before normalization (Barrière et al., 2008, 2010), which demonstrated that initial symmetry restoration was independent from descending influences. In contrast, other studies in animals (Singh et al., 2011) and humans (Edgerton et al., 2001) tend to suggest that recovery from unilateral SCI depends strictly on an interplay between contralesional descending pathways and local sensory feedback to enhance depolarization in spinal neurons below the lesion at least primarily; thereafter, when spinal circuits retrieve their ability to respond to descending commands without additional excitation, reflexes would be down-regulated (Little et al., 1999). Experiments in cats showed that restoration substantially relied on the asymmetrical integration of local reflexes, notably cutaneous-mediated reflexes (Helgren and Goldberger, 1993; Frigon et al., 2009) that were enhanced in the ipsilesional lumbar hemicord (Gossard et al., 2015). Similar implication of cutaneous inputs was also reported from thoracic hemisection experiments in the chick (Muir and Steeves, 1995; Muir et al., 1998), and conditioning the control of H-reflex amplitude on the ipsilateral side reduces motor asymmetry in rats, monkeys and humans (Wolpaw and Lee, 1989; Chen et al., 2006; Thompson and Wolpaw, 2015). In contrast, prior ankle denervation prevented symmetrical posturo-locomotor recovery in SCI cats (Carrier et al., 1997). All these results support a pivotal role for local sensory integration in the recovery of functional motor symmetry. Furthermore, mutant mice lacking proprioceptive feedback from muscles showed no recovery after a thoracic hemisection, and this was associated with a restricted and inaccurate reorganization of contralesional descending pathways below the lesion (Takeoka et al., 2014). This latter study thus suggests that if an interplay between descending pathways and local sensory feedback is required for the accurate reorganization of spinal networks after unilateral SCI, local sensory information likely orchestrates this interplay.

Plasticity is not limited to the restoration of hindlimb CPG operation. After a cervical hemisection, modifications were found in the unaffected forelimb circuitry that compensated for the motor deficits in the harmed forelimb in adult rats (Khaing et al., 2012), and the bilateral propriospinal coordination between cervical and lumbar networks (respectively controlling forelimbs and hindlimbs) is changed in order to bilaterally restore the anteroposterior interlimb coordination in cats (Côté et al., 2012). Although significant posturo-locomotor recovery occurs, motor anomalies may nevertheless persist after incomplete SCI and subsequent training. In adult rats, for instance, persistent overlap between flexor and extensor muscle activities was reported and proposed to result from the existence of an asymmetry in descending pathways (Kaegi et al., 2001). However, the authors also reported a simultaneous improvement of stance and body support in lesioned animals, which suggested that what appeared to be an incomplete recovery might in fact consist of a compensatory motor adaptation to the loss of half of the descending motor commands. In fact, global reorganization of muscle activation pattern was similarly described where axial muscles below the lesion were found to participate more during pelvis and hindlimb movements in trained SCI rats (Giszter et al., 1998), as well as during phase shift in walking patients (Scivoletto et al., 2007). Such axial muscle plasticity further raises the question of why axial networks seem to be less affected by unilateral central lesions. An explanation may be found in the particular organization of descending commands onto these networks where pyramidal excitatory commands onto dorsal MNs are mainly relayed by RS neurons and inhibitory commands by spinal inhibitory INs located in the two hemicords. Such bilaterally distributed organization may account for the axial system greater resilience to unilateral pyramidectomy in cats (Galea et al., 2010).

Because both ascending (principally sensory) and descending (motor) pathways are affected by SCI protocols, series of experiments with restrained spinal lesion were designed with the aim of clarifying the relative participation of different pathways in SCI acute effects and subsequent recovery. Hence, a joint unilateral destruction of dorsolateral funiculus (corticospinal and reticulospinal tracts) and ipsilateral dorsal columns (ascending sensory tract) had no long-lasting impact on posture and locomotion in adult rats. In contrast, it produced strong deleterious effects when combined with ventrolateral funiculus (principal reticulospinal tract) interruption (Loy et al., 2002), confirming the major role of RS neurons in the control of downstream motor networks. Such a role was further supported by the observation in rats that post-SCI locomotor recovery necessitated the preservation of at least part of spinal ventral and/or lateral white matter (Schucht et al., 2002). Furthermore, increased crossing of contralesional RS fibers and increased number of connections on local propriospinal INs were reported to develop below the hemisection site, together with increased responses of these RS neurons to stimulation of the contralesional MLR (Rangasamy, 2013; Zörner et al., 2014; Engmann et al., 2020). This suggests that plasticity in brainstem motor command circuitry occurs to compensate the functional loss of a half of command neurons while spared RS neuron projections reorganize to restore functional symmetry in projection pathways below the lesion (Figure 3A).

In contrast, after unilateral spinal hemisection no significant changes were globally found in cortico- and vestibulospinal tracts, and only a small tendency of increased rubrospinal fiber crossing was reported (Zörner et al., 2014). However, selective spinal lesion targeting the decussating corticospinal tract on one side induced motor deficits that recovered rapidly in rats (Bazley et al., 2012). This was paralleled by the induction of new connections between corticospinal terminals and long-projecting propriospinal INs that connected lumbar MNs (Bareyre et al., 2004), the outgrowth of ipsilaterally projecting corticospinal fibers into sub-lesion cord segments, as well as sprouting of ipsilesional sensory afferents within the ventral cord territories that lost their corticospinal inputs (Figure 3A; Bareyre et al., 2002; Brus-Ramer et al., 2007; Tan et al., 2012). The establishment of new corticospinal connections and subsequent motor recovery are prevented in mutant mice expressing inhibitory DREADDs in spinal INs, demonstrating that this process depends on the presence of activity in local INs, likely induced by segmental sensory afferent activation of NMDA receptors (Bradley et al., 2019). Consecutive to sensory-motor cortex unilateral stroke (which causes corticospinal neuron degeneration) hindlimb stretch reflexes were found to be enhanced and some plasticity genes (e.g., Grin2a/DLg4 and Tgfb1) inversely regulated in adult rat lumbar segments (Zhang et al., 2020). This study suggests that one important function for sensory-motor cortex neurons projecting to the spinal cord is to regulate plastic adaptation of local spinal reflexes during ongoing posturo-locomotor activity. Indeed, phase-dependent depression of reflex loops was found to be reduced following incomplete SCI and able to recover, in a neurotrophic factor-dependent manner (Côté et al., 2011), in response to spared corticospinal pathway stimulation (Brus-Ramer et al., 2007; Petrosyan et al., 2020) or sensory-motor training (Phadke et al., 2009; Côté et al., 2011). Corticospinal damage-induced spinal reorganization acutely leads to spinal hyperreflexia, which has been considered as maladaptive (Tan et al., 2012), though such hyperreflexia may participate actively in the process of posturo-locomotor recovery by increasing the overall excitability of spinal networks below the lesion site (see below).

Dorsal hemisection, which suppress half of sensory inputs ascending to brain centers, has dramatic acute effects on locomotion, affecting notably paw placing during skilled walking, until sensory afferent reorganization within the ipsilesional hemicord occurs and allows functional recovery. In mice with such lesions, ascending dorsal root neurons increased branching below the lesion site and, in particular, connected propriospinal INs projecting into the cuneate nucleus, constituting so a detour circuit to brainstem centers (Granier et al., 2020), and beyond, to the primary sensory-motor cortex (as afore suggested in rats and monkeys; Kaas et al., 2008). Functional restoration of bilateral tactile and proprioceptive sensation occurred in these animals within weeks, demonstrating the exceptional ability of sensory systems to compensate for selective alterations of their central pathways. In contrast, adaptation of central motor commands to unilateral sensory loss noticeably takes much longer to reorganize. Indeed, it was shown in macaque monkeys that a unilateral dorsal rhizotomy at the cervical level (which completely removed sensory inputs from one forelimb) triggered a complete loss of grasp in the corresponding forelimb that persisted over several months. With time, functional recovery nevertheless occurred and was paralleled by the anatomical reorganization of corticospinal projections into cervical segments from both the primary somatosensory and motor cortices (Darian-Smith et al., 2013; Nardone et al., 2013); however, while the former displayed clear retraction from the cervical hemicord, the latter exhibited abnormal sprouting in the ipsilesional dorsal horn (Darian-Smith et al., 2013; but see also Darian-Smith et al., 2014). These works demonstrate that a unilateral sensory deprivation causes rearrangement of different corticospinal projections, and further suggest a distinct implication of each corticospinal tract in motor recovery that remains unexplained.

Whereas the effects of cord hemisection on directly lesioned tissues are well documented (Ahuja et al., 2017), the cellular impacts on downstream motor networks in relation with the observed behavioral consequences is still not clear, and some aspects have not been explored to date. For example, the astrocyte role in SCI-induced formation of glial scars (astrocytic reaction) has been analyzed extensively at the lesion level itself (Okada et al., 2018). Surprisingly, even though they are known to regulate activity of locomotor networks in physiological conditions (e.g., Mu et al., 2019) the potential implication of astrocytes in post-SCI motor adaptation remains undocumented. In contrast, numerous investigations in various species have analyzed the neuronal mechanisms involved in motor restoration. In injured lampreys, spinal neurons below the lesion exhibited increased resting membrane potential as well as increased input resistance that, combined with synaptic efficacy enhancement, strongly improved neuronal excitability (Cooke and Parker, 2009). In adult rats with cervical cord hemisection, Gonzalez-Rothi and colleagues reported the transient development of muscular atrophy in the corresponding forelimb (Gonzalez-Rothi et al., 2016), which could have suggested a deficit in motor command due to dramatic and maladaptive rearrangement of spinal motor and/or premotor neurons. But, using pseudorabies viruses during the same period of time in order to track putative added/subtracted connections, the authors did not observe any obvious changes in first-order premotor IN-MN circuitry of the cervical cord (Gonzalez-Rothi et al., 2015), suggesting that no dramatic premotor network reorganization occurred that could explain muscle atrophy-related reduced motoneuronal activity, nor its recovery with time. Furthermore, it was shown in mice that genetically identified V2a (Shox2) INs, which are putative constituents of the lumbar CPG, conserved their intrinsic resting membrane properties after SCI (Husch et al., 2012). However, these INs exhibited increased excitability due to enhanced efficacy of both local sensory and descending monoaminergic input synapses, as well as more intense expression and increased sensitivity of 5-HT_2B/_2C receptors (Figure 3B; Husch et al., 2012; Garcia-Ramirez et al., 2021). Noteworthy, these serotonin receptors facilitate synaptic input-evoked responses of spinal neurons, likely by modulating calcium channels sustaining plateau potentials (Perrier et al., 2013). Taken together, it seems that spinal INs are only minimally affected by spinal hemisection, suggesting a subsequent minor role in motor symmetry restoration. Nevertheless, because these INs integrate inputs from both local sensory afferents and descending projections from brainstem command centers, which both exhibit deep adaptation (see above), it is likely that plasticity at IN input synapses would allow more efficient network operation. Hence, increasing inhibition from commissural INs would participate in the restoration of bilateral coordination, while enhanced ipsilateral inhibition would participate in reducing MN hyper-excitability and in shaping new adapted coordination.

In contrast, diverse modifications were reported in MNs in rodents which underwent SCI (Figure 3B). Sub-lesion MNs were characterized by increased expression of sustained Na/Ca-dependent plateau potentials (Li et al., 2004; Heckman et al., 2005) and NMDA receptors (Gransee et al., 2017), both increasing MN responses to excitatory synaptic inputs. Moreover, structural alterations of motoneuronal input synapses favoring MN excitability were reported in the lamprey (Cooke and Parker, 2009), and a reduction of potassium conductances occurs in ipsilesional leg MNs after unilateral interruption of the corticospinal tract in human patients (Jankelowitz et al., 2007). Rodent sub-lesional MNs were also shown to express less chloride transporters KCC2 at their plasma membrane, which diminishes the impact of the inhibitory inputs they integrate (Boulenguez et al., 2010). Moreover, SCI triggered a decreased expression of membrane receptors to inhibitory neurotransmitters GABA and glycine in downstream MNs (Sadlaoud et al., 2010, 2020), which modifies motoneuronal integration of inhibitory inputs (de Leon et al., 1999). Globally, this combination of compensatory changes in motoneuronal properties results in a dramatic increase in MN excitability. One may ask whether changes in MN dendritic morphology could happen after a cord hemisection, since modified dendritic arbors were reported in lumbar MNs from rats with complete thoracic transection (Gazula et al., 2004). Noteworthy, phrenic MNs do not exhibit morphology changes after a cervical hemisection, whereas it is the case after complete activity blockade with tetrodotoxin (Mantilla et al., 2018). Thus, dendritic morphology likely is directly linked to overall activity rather than to precise input synapses. As a consequence, because cord hemisection globally increases neuronal activity we may not expect significant dendritic morphology changes in sub-lesion MNs. Interestingly, it has been reported from mice that acute thoracic hemisection reduced the number of input synapses onto MNs located upstream of the lesion (Goldshmit et al., 2008), an effect that was reversed by motor training. This latter result demonstrates that MNs throughout the spinal cord are subject to plasticity in response to localized SCI and further support the idea that global symmetry restoration in posturo-locomotor functions likely depends on cellular modifications spreading all over central motor circuits.

Either the large majority of studies until now have focused on MN properties and very few on IN ones (yet, see the demonstration of IN neurotransmitter switching in the autonomic system after spinal lesion; Hou et al., 2016) or the principal cellular target of unilateral SCI-triggered plasticity is the MN. This gap in knowledge questions the inhibitory network modification that has been proposed to account for the alteration of sensory integration in humans with incomplete SCI (Knikou, 2012), which could finally result also/instead from changes in motoneuronal integrative processes below the lesion as shown in adult rats (Sadlaoud et al., 2020). Further investigations at the cellular level will be necessary to unravel the relative participation of network INs and output MNs in motor recovery after unilateral SCI.

Early events that evoke vestibular imbalance are followed, with variable timelines depending on the considered species, by further plasticity mechanisms leading to a nearly full recovery of static head and trunk posture while dynamic deficits are never fully restored. Supposedly, these mechanisms constitute the substrate for the so-called ‘vestibular compensation,’ taking place in every different structure mentioned above and resulting in the re-equilibration of CVN neuronal resting discharge on the two sides (Ris et al., 1997; Vibert et al., 1999b). At the CVN level, the restoration of balance results principally from modifications in the connectivity and efficacy of both sensory and commissural vestibular connections associated with changes in intrinsic properties of secondary vestibular neurons on the side of the lesion (Figure 4B). Such a combination of pathway reorganization and post-lesional cellular plasticity was suggested to account for the correction of static syndromes (see Lambert and Straka, 2012).

The specific interruption of semi-circular canal signals on one side early after birth delayed but did not prevent acquisition of motor abilities in rats (Geisler et al., 1997). However, no clear modifications of the temporal organization of locomotion were observed while trunk stability remained non-optimal for a longer time, thus suggesting that retardation might result from a deficit in dynamic coordination between locomotion and posture. Whether such coordination deficits resulted exclusively from the persistent default in descending commands or also from a lack of spinal network adaptation was not investigated. In fact, only few studies have analyzed the impact of a UVD on the subsequent development of spinal neural networks involved in the control of posture and locomotion, and the clearer results come from studies in the aquatic anuran Xenopus laevis (Figure 5).

In the larval Xenopus, as in all vertebrates, a UVD causes a loss of balance between ipsi- and contralesional CVNs; however, contrary to the other animal models, this imbalance is never compensated for and even subsists during metamorphosis to be similarly present in the post-metamorphic frog (Lambert et al., 2008, 2013; Beyeler et al., 2013). Interestingly, UVD effects on posture are identical whether it is performed before (tadpole stage) or after metamorphosis (juvenile stage), with a typical head twist and trunk curvature toward the lesioned side that remains uncorrected, likely because of the lack of ground-related sensory feedback in this purely aquatic animal. Performed at either developmental stage, UVD also similarly evoked locomotor deficits, in the form of a constant rolling behavior toward the lesion side (Beyeler et al., 2013). However, whereas swimming remained impaired in animals lesioned after metamorphosis, those lesioned before metamorphosis recovered near-normal locomotion after metamorphosis completion (Beyeler et al., 2013) whilst brainstem descending commands stayed permanently unbalanced (Lambert et al., 2013). Both in vivo and in vitro analyses of the spinal network responsible for dynamically coordinating posture and propulsion during swimming in post-metamorphic frogs demonstrated dramatic alteration in juveniles that have been lesioned before metamorphosis, compared to control animals or juveniles in which UVD had been performed after metamorphosis (Beyeler et al., 2013). Whereas both control (Beyeler et al., 2008) and juveniles vestibulo-lesioned after metamorphosis possessed functionally symmetrical spinal posturo-locomotor networks (Figures 5A,B), juveniles that were lesioned before metamorphosis exhibited an asymmetrical network, characterized by the functional loss on the lesioned side of the lumbo-thoracic coordination subduing axial postural muscles to the appendicular propulsive system (Figure 5C). Biomechanical models demonstrated that this simple alteration was sufficient to counteract the unbalanced brainstem descending commands evoked by the UVD. Indeed, UVD-induced disequilibrium consisted of a relative increase of the commands descending into the ipsilesional side of the cord, notably because of the loss of ipsilesional tangential neurons that normally project into the contralesional hemicord (Lambert et al., 2013). Hence, the ‘developmental suppression’ of ipsilesional lumbo-thoracic coordination results, during swimming, in the loss of propriospinal excitation onto postural MNs specifically on the lesion side, while contralateral MNs still receive this ascending excitation. As a result, during swimming the increased descending command on the ipsilesional side is compensated for by the absence of ascending propriospinal excitation, while the reduced descending command on the contralesional side is compensated for by the persistence of (and thus, relative to the other hemicord, increase in) ascending propriospinal excitation originating from the lumbar CPG. Altogether, these two opposite asymmetries lead to re-balancing activity in the spinal postural system during locomotion and allow normal (non-rolling) swimming in juveniles that underwent a UVD before metamorphosis.

Whether such spinal reorganizations are specific to Xenopus metamorphosis or occur similarly in other vertebrate species remains to be determined. Although very few investigations have been made in this direction, several indirect results tend to indicate that UVD-triggered spinal network adaptations may constitute a general feature in a majority of vertebrates. Hence, adult UVD rats that display severe postural defects at rest and dramatically unstable locomotion at low speed of walking, exhibit a more stable body balance with increased locomotor velocity (Rastoldo et al., 2020). Similarly, a running UVD dog shows less deviation from the normal path and less imbalance than during walking (Brandt et al., 1999). Interestingly, the exact same observations were made in patients with acute vestibulopathy in which walking strides were more regular and body balance more accurate at high velocity than at low speed (Brandt et al., 1999; Schniepp et al., 2012; reviewed in Akay and Murray, 2021). Two explanations could account for such a ‘speed-dependence’ of posture and gait in UVD subjects. First, this could result from decreasing the weight of vestibulospinal control onto spinal networks during high velocity locomotion as proposed by Akay and Murray (2021). Indeed, a study in healthy human subjects showed that gait deviation induced by a vestibular galvanic stimulation was reduced by half during running compared to walking, suggesting that higher locomotor speed exerted stronger inhibitory control on descending vestibulospinal commands (Jahn et al., 2000). In this study, however, galvanic stimulation mimicked sudden shifts of head position while locomotion was already ongoing, generating disrupting vestibular signals thus superimposed on still symmetrical tonic descending influences from gravity detectors. Of course, it is totally different in subjects with acute vestibular imbalance, where the basal tonic influence is no more symmetrical. An alternative explanation would arise from the possibility that, as demonstrated in juvenile Xenopus (Beyeler et al., 2008), the spinal posturo-locomotor network is intrinsically able to organize postural controls that are dynamically adapted to the ongoing propulsive motor program. In this context, locomotion and dynamic control of posture would be principally based on highly automated programs organized by spinal networks coordinated through propriospinal interconnections and under control of both tonic and dynamic supraspinal modulation (e.g., Olechowski-Bessaguet et al., 2020). The relative weight of propriospinal coordination vs. descending commands would then determine how the dynamic control of posture would be achieved (Beyeler et al., 2013): stronger is the spinally generated motor command, lower is the relative weight of dynamic descending inputs. Whether such balance is purely orchestrated at the spinal synaptic level or rather/additionally involves direct modulation of sensory inputs at the supraspinal level (Le Ray et al., 2010; Chagnaud et al., 2015; Straka et al., 2018) still has to be determined. Thus, running may rely much less on dynamic descending inputs than slow walking does. Yet, after UVD this would require precise alterations of the spinal posturo-locomotor network taking into account the persistent disequilibrium in tonic descending influences. However, contrary to what was described in the metamorphosing Xenopus where propriospinal adaptation result from developmental processes, it is likely that such alterations in mammals would rather involve plastic adaptations closer to those consecutive to a unilateral SCI. The existence of such UVD-triggered spinal reorganization, whether it is developmental or adaptive, confronts the classical conclusions about UVD effects on the control of posture (e.g., Vidal et al., 1993) which were drawn from studies restricted to the control of the so-called ‘static’ posture.