The neural control of interlimb coordination during mammalian locomotion

Abstract

Neuronal networks within the spinal cord directly control rhythmic movements of the arms/forelimbs and legs/hindlimbs during locomotion in mammals. For an effective locomotion, these networks must be flexibly coordinated to allow for various gait patterns and independent use of the arms/forelimbs. This coordination can be accomplished by mechanisms intrinsic to the spinal cord, somatosensory feedback from the limbs, and various supraspinal pathways. Incomplete spinal cord injury disrupts some of the pathways and structures involved in interlimb coordination, often leading to a disruption in the coordination between the arms/forelimbs and legs/hindlimbs in animal models and in humans. However, experimental spinal lesions in animal models to uncover the mechanisms coordinating the limbs have limitations due to compensatory mechanisms and strategies, redundant systems of control, and plasticity within remaining circuits. The purpose of this review is to provide a general overview and critical discussion of experimental studies that have investigated the neural mechanisms involved in coordinating the arms/forelimbs and legs/hindlimbs during mammalian locomotion.

proper coordination of the four limbs is a requirement of terrestrial locomotion in mammals, including humans. A precise yet flexible control of interlimb coordination allows an animal to maintain dynamic stability in a continuously changing environment, when changing speed, or when transitioning from one gait pattern to another. It also allows the arms/forelimbs to perform movements that are independent of the legs/hindlimbs during locomotion, such as reaching or grasping. An effective coordination of the limbs is achieved through a complex interplay between spinal circuits that generate the basic pattern of locomotion, somatosensory feedback that informs the central nervous system (CNS) of changes within the body and in the environment, and supraspinal structures that regulate posture and volitional aspects of locomotion. The purpose of this review is to provide a general overview and critical discussion of the neural structures and pathways involved in coordinating the arms/forelimbs and legs/hindlimbs during mammalian locomotion. I start some of the terminology used to define interlimb coordination, followed by the types of gait patterns that can be produced during quadrupedal locomotion. I then discuss the evidence showing that each limb is controlled by its own spinal locomotor network, followed by a discussion of how these networks are coordinated by spinal mechanisms, somatosensory feedback from the limbs, and various descending pathways from supraspinal structures. The neural control of interlimb coordination is then summarized in a schematic representation of the different control systems that can interact with the spinal locomotor networks. I then discuss studies that have investigated the coordination between the forelimbs and hindlimbs during locomotion in mammalian models of spinal lesions. The work in animal models is then followed by a discussion of the neural control of the arms and legs in healthy humans and in people with spinal cord injury (SCI). The focus of the present review is on the neural mechanisms coordinating arm/forelimb and leg/hindlimb locomotor movements. For reviews on the neural mechanisms involved in coordinating the left and right sides of the spinal cord at segmental levels (e.g., the two hindlimbs), I refer the reader to other articles that have specifically covered this topic (Bernhardt et al. 2013; Butt et al. 2002; Butt and Kiehn 2003; Kiehn 2016; Shevtsova et al. 2016).

Some Notes on Terminology

In this review, I use “interlimb coordination” as a general term to denote the coordination between the four limbs. The term “left-right coordination” is used to define the coordination between the two limbs of the same girdle (i.e., left forelimb-right forelimb or left hindlimb-right hindlimb), also sometimes referred to in the literature as homologous coupling. “Forelimb-hindlimb coordination” or “arm-leg coordination” is used to represent the coordination between the arms/forelimbs (pectoral/shoulder girdle) and legs/hindlimbs (pelvic girdle). Forelimb-hindlimb or arm-leg coordination is further divided into “homolateral coordination,” which refers to the coordination between the two limbs on the same side of the body (i.e., left forelimb-left hindlimb or right forelimb-right hindlimb), and “diagonal coordination,” to represent the coordination between a hindlimb and the forelimb located on the contralateral side (i.e., left forelimb-right hindlimb or right forelimb-left hindlimb). These are useful terms to explain the different types of gait patterns that can be produced by a quadruped during locomotion and the potential pathways involved in coordinating the four limbs, as well as how interlimb coordination and its control are altered after SCI. The term “spinal” or “spinal animal” (e.g., spinal cat) is used to refer to animals with a complete spinal transection.

Different Types of Gait Patterns

The flexibility in the control of interlimb coordination can be appreciated by the different types of gait patterns that can be produced during quadrupedal locomotion. Gait patterns can be broadly divided into symmetrical and asymmetrical gaits, which are defined by the period of time that a limb is contacting the ground/surface, termed the duty cycle, and the pattern of coordination between the limbs (Hildebrand 1965, 1989). Symmetrical gaits occur when contacts between the two forelimbs and between the two hindlimbs are evenly spaced in time (Abourachid 2003; Bellardita and Kiehn 2015; Cartmill et al. 2002; Hildebrand 1965, 1967, 1976; Lemieux et al. 2016; Patrick et al. 2009; Vilensky and Larson 1989), or in other words, when the right forelimb contacts the ground at 50% of the cycle between two successive left forelimb contacts. The same applies to the hindlimbs. Symmetrical gaits include the walk, trot, and pace. The walk is most commonly found at the slowest speeds and is characterized by a pattern where three or four limbs are simultaneously contacting the ground due to the long stance phases of individual limbs (>70% of total cycle). The trot is characterized by simultaneous forward movements of the diagonal limbs, whereas the pace is characterized by simultaneous forward movements of homolateral limbs.

Another way to express symmetrical gait patterns is through the sequence of paw contacts, or footfall patterns, which can be separated into lateral or diagonal sequences (Cartmill et al. 2002; Hildebrand 1976; Lemelin et al. 2003; Vilensky and Larson 1989). In a lateral sequence, the footfall pattern proceeds in the following order of contacts: right hindlimb, right forelimb, left hindlimb, and left forelimb. The lateral sequence is adopted preferentially during overground locomotion by a number of quadrupedal mammals, including cats and dogs, as well as by human infants and adults during crawling (Patrick et al. 2009, 2012; Righetti et al. 2015). In a diagonal sequence, the footfall pattern proceeds with right hindlimb, left forelimb, left hindlimb, and right forelimb contacts. Many nonhuman primates use a diagonal sequence during overground locomotion, although several primate species can also use a lateral sequence (Cartmill et al. 2002; Hildebrand 1967, 1976, 1989; Lemelin et al. 2003; Vilensky and Larson 1989).

Despite the preferential use of one sequence over the other, quadrupeds can also switch to the other sequence depending on task demands. For instance, when the forelimbs and hindlimbs stepped on two different belts and at different speeds on a transverse split-belt treadmill, both the lateral and diagonal sequences could be observed in adult cats (Thibaudier et al. 2013). Hildebrand proposed that increasing the duration of the hindlimb stance phase relative to the forelimb stance phase (i.e., the ratio of their duty cycles) could transform a lateral sequence to a diagonal sequence, or vice versa (Cartmill et al. 2002; Hildebrand 1976). It also has been proposed that the type of sequence selected is dependent on body weight distribution (Rollinson and Martin 1981; Vilensky and Larson 1989). During quadrupedal locomotion, cats and dogs bear a greater percentage of their bodyweight on their forelimbs and preferentially use a lateral sequence, whereas many nonhuman primates bear more weight on their hindlimbs and primarily use a diagonal sequence (Vilensky and Larson 1989). By using a quadrupedal robot with a central pattern generator (CPG)-based control of each limb that receives local sensory feedback, it was shown that a switch from a lateral to a diagonal sequence occurred in the gait pattern of the robot by increasing the load on the hindlimbs (Owaki et al. 2012). Hindlimb loading increased the duration of the hindlimb stance phase relative to forelimb stance phase during robot locomotion and, as predicted by Hildebrand (1976), transformed a lateral sequence to a diagonal one. However, a change in the ratio of hindlimb and forelimb stance phases to explain a switch in sequence does not hold true for all species tested, suggesting that other factors are involved, such as a need to maximize dynamic stability (Cartmill et al. 2002).

Asymmetrical gaits occur when homologous limbs at the pectoral/shoulder or pelvic girdles begin to move toward an in-phase pattern or in synchrony. Asymmetrical gaits include the bound or gallop. The bound is a form of gait where homologous limbs perform synchronous movements with the hindlimbs pushing off together and the forelimbs landing simultaneously. The short stance phases allow for an aerial phase where all four limbs are off the ground. The gallop, which is found from intermediate to the fastest speeds, is also characterized by an aerial phase of all four limbs, but contrary to the bound, both pairs of homologous limbs are not in perfect synchrony. Mice exhibit a half-bound gallop where the hindlimbs move in synchrony while the forelimbs are out of phase (Bellardita and Kiehn 2015). In other types of gallop, one limb leads its homolog.

The type of gait expressed in quadrupedal mammals seems to depend on the recruitment of specific populations of spinal neurons. For instance, in mice lacking V0 neurons, a type of ventrally derived progenitor neuron that gives rise to spinal commissural interneurons (Goulding 2009), the only gait that could be expressed was bound, even at slow locomotor speeds where this gait is not normally expressed (Bellardita and Kiehn 2015; Talpalar et al. 2013). Genetic ablation of only a subset of V0 neurons, the excitatory ventral V0 population (V0_V), abolished the capacity for trot but left the ability to produce walk, bound, and gallop unaffected in mice (Bellardita and Kiehn 2015). However, although bound and gallop were present, they lacked an aerial phase due to longer stance durations. Moreover, Bellardita and Kiehn (2015) showed that walk, trot, and bound occurred at mostly nonoverlapping frequencies/speeds of locomotion, with abrupt transitions between gait patterns, whereas gallop overlapped with trot and bound. They proposed that their results were consistent with a modular organization of spinal circuits, with different ensembles of neurons producing specific gait patterns.

In summary, mammalian quadrupeds are capable of performing different types of gait patterns to meet task requirements, highlighting the flexibility of the systems coordinating the limbs. Although the neural mechanisms involved in producing a given type of gait in mammals remain largely unknown, the activation of different subsets of spinal neurons, or modules, appears to be involved.

Each Limb Is Controlled by a Separate Spinal Network

How do the limbs produce locomotor movements? At the core of the neural control of locomotion is the spinal cord, which directly controls all movements of the trunk and limbs. The spinal cord is the main integrative center of the CNS for motor control, receiving inputs from peripheral receptors and from supraspinal structures to directly control and adjust movements. Spinal circuits, by themselves, can generate simple motor behaviors, such a withdrawal reflexes, or more complex ones, such as locomotion. Indeed, it is now well established that the basic locomotor output is controlled by a network of spinal neurons, the central pattern generator (CPG) (reviewed in Frigon 2012; Grillner 1981; Grillner and El Manira 2015; Grillner and Jessell 2009; Guertin 2009; Kiehn 2016; McCrea and Rybak 2008; Rossignol et al. 2006; Rossignol and Frigon 2011; Stuart and Hultborn 2008; Zehr et al. 2016).

Since the pioneering work of Sherrington (1910) and Brown (1911, 1914) in the early part of the 20th century, methods to study spinal mechanisms of locomotor control in mammals have become more sophisticated. One particularly fruitful approach has been to study locomotor-like activity in curarized decerebrate or decorticate animals, mainly cats (Andersson et al. 1978; Grillner 1981; Grillner and Zangger 1979; Jankowska et al. 1967a, 1967b; Orsal et al. 1990), or in isolated or semi-isolated spinal cord preparations in neonatal rats or mice (Cazalets et al. 1992; Cowley and Schmidt 1997; Kjaerulff and Kiehn 1996; Whelan et al. 2000). The locomotor-like output is termed “fictive locomotion” because although the pattern of motor activity recorded from muscle nerves or ventral roots is locomotor-like (e.g., flexor-extensor and left-right alternation), it is not real locomotion due to the absence of movement. Advantages of this approach are that phasic sensory feedback from peripheral mechanoreceptors is absent and the modulation of the locomotor-like pattern can be modulated pharmacologically, electrically, or mechanically. In fictive preparations, the rhythm is also referred to as “locomotor-like” because other types of rhythmic motor patterns can also be evoked and recorded, such as scratch (Arshavsky et al. 1978; Frigon and Gossard 2010; Perreault et al. 1999; Power et al. 2010) and fast paw-shake (Pearson and Rossignol 1991). The presence of different rhythmic motor behaviors suggests the existence of different spinal CPGs and/or different possible configurations of the spinal network (discussed in Frigon 2012). For instance, in the chronic spinal cat, hindlimb locomotion is abolished with a spinal transection at mid-lumbar levels, whereas the fast paw-shake rhythm persists (Langlet et al. 2005), indicating that the two rhythms are generated by specialized circuits. In vitro, different rhythms can be elicited by changing the type of neurochemicals that are bath-applied to the spinal cord (Cowley and Schmidt 1994, 1997; Kiehn and Kjaerulff 1996). The presence of different spinally generated rhythms emphasizes the importance of considering the type of pattern that is being studied when interpreting data.

Distinct central pattern generators for the forelimbs and hindlimbs.

Pharmacologically evoked fictive locomotor-like rhythms of the fore- and hindlimbs have been recorded in mammalian models, such as cats and rabbits, following an acute complete spinal transection at upper cervical levels (high spinal animals) (Viala and Vidal 1978; Zangger 1981) or from cervical and ventral roots of isolated spinal cord preparations in neonatal rodents (Ballion et al. 2001; Juvin et al. 2005, 2012). This indicates that the neuronal circuits producing forelimb and hindlimb locomotion are located within the spinal cord. In studies where fore- and hindlimb nerves or cervical and lumbar roots were recorded simultaneously, the rhythms at both levels were generally of similar frequency (Ballion et al. 2001; Juvin et al. 2005, 2012; Orsal et al. 1990; Viala and Vidal 1978). To show that the cervical and lumbar rhythms were coordinated, the phasing between bursts at both levels was measured and analyzed for statistical significance using circular statistics (Ballion et al. 2001; Juvin et al. 2005, 2012; Orsal et al. 1990). In these studies, data points were clustered around specific values, indicating strong interactions between cervical and lumbar CPGs.

The evidence that the forelimbs and hindlimbs are controlled by functionally and anatomically distinct spinal locomotor CPGs was first obtained from lesion studies. For instance, fictive locomotor-like rhythms have been recorded in forelimb (Arshavsky et al. 1986; Gödderz et al. 1990; Saltiel and Rossignol 2004; Viala and Vidal 1978; Yamaguchi 2004) and hindlimb motor nerves (Fedirchuk et al. 1998; Frigon and Gossard 2009; Grillner and Zangger 1979; Kiehn et al. 1992; Meehan et al. 2012) following a complete spinal transection at thoracic levels in a variety of decerebrated mammals. After spinal transection or synaptic blockade at thoracic levels, the cervical and lumbar rhythms are physically and/or functionally uncoupled, with no apparent phasing between the activities recorded at cervical and lumbar levels (Ballion et al. 2001; Juvin et al. 2005). When this occurs, the cervical rhythm can be slower (Ballion et al. 2001; Juvin et al. 2005; Viala and Vidal 1978) or slightly faster (Juvin et al. 2005) than the lumbar rhythm. The hindlimb locomotor-like rhythm can also be faster when evoked by electrical stimulation of the brain stem (Rossignol et al. 1993). Thus it does not appear that cervical or lumbar CPGs have inherently faster rhythmogenic properties, and the observation that one rhythm has a higher frequency than the other is most likely preparation dependent (e.g., supraspinal drive, pharmacology, transection/blockade level).

Critical elements of the forelimb CPGs were found to be located at low cervical/upper thoracic spinal segments (Ballion et al. 2001; Yamaguchi 2004), whereas those of the hindlimb CPGs were identified at upper to mid-lumbar spinal segments (Cazalets et al. 1995; Langlet et al. 2005; Marcoux and Rossignol 2000). Although some spinal segments are critical for the generation of forelimb and hindlimb locomotion, it was shown that rhythmogenic capabilities are distributed along the length of the spinal cord (Cowley and Schmidt 1994, 1997; Delivet-Mongrain et al. 2008; Hägglund et al. 2013; Kjaerulff and Kiehn 1996). Functionally, this distributed rhythmogenic capacity allows various spinal segments located between cervical and lumbar enlargements to locally activate descending and ascending pathways that communicate between the locomotor CPGs controlling the fore- and hindlimbs. An advantage of such a distributed system is that the fore- and hindlimb CPGs can be more easily uncoupled for independent use of the forelimbs, which is particularly important for many mammals.

Distinct central pattern generators for the left and right sides.

The evidence that locomotor movements of the two forelimbs or two hindlimbs are generated by different spinal locomotor CPGs is derived from the following observations. First, locomotor studies in spinal cats that were performed on a split-belt treadmill, where the stepping speed of each hindlimb can be independently controlled, have shown that the left and right hindlimbs can step at different rhythms (Forssberg et al. 1980; Frigon et al. 2013, 2017). In other words, the hindlimb stepping on the faster belt can take more steps than the one on the slower belt. Ratios up to 5:1 (fast-slow hindlimbs) have been observed. A similar phenomenon is observed in human infants during split-belt locomotion (Yang et al. 2005). Although human adults have difficulty performing an unequal number of steps between the legs during split-belt locomotion, they can perform forward stepping with one leg and backward stepping with the other (Choi and Bastian 2007). These results are consistent with autonomous locomotor networks for the left and right sides.

Second, lesions along the midline of the spinal cord have also been used to anatomically uncouple the left and right spinal locomotor CPGs. For instance, Kudo and Yamada (1987) used a spinal cord preparation (L₁–L₆ segments only) with the hindlimbs attached and evoked locomotor-like activity by bath applying N-methyl-d,l-aspartate (NMA) in neonatal rats (postnatal days 0–3). In some preparations, they separated the left and right sides of the spinal cord by cutting it along the midline. They showed that each half of the spinal cord could produce its own locomotor rhythm, albeit at a slower frequency than when both halves were connected. In another study, Kato (1990) performed a lateral spinal hemisection combined with a midline lesion caudal to the hemisection to anatomically isolate the neural network controlling one hindlimb. The isolated hindlimb regained the ability to step a few days after the other three limbs recovered. However, its phase relationship with the other limbs was inconsistent on a step-by-step basis, indicating that the isolated hindlimb was independent of the other three, most likely driven by local somatosensory feedback.

In summary, experimental studies in mammals have shown that locomotor movements of the four limbs are generated by distinct spinal neuronal networks, one for each limb. The following sections discuss the neural mechanisms coordinating the fore- and hindlimbs during locomotion in mammals.

Neural Structures and Pathways Involved in Forelimb-Hindlimb Coordination

Before discussing the potential neural structures and pathways controlling forelimb-hindlimb coordination, it is important to point out that mechanical linkages between the limbs and trunk also play a major and essential role in stabilizing coordinated multilimb movements. In turn, the CNS is continuously informed of the mechanical state of the body as it interacts with the environment via somatosensory feedback from the periphery. How interactions between neural and mechanical factors shape coordinated movements is one of the major challenges in neurobiology today (Nishikawa et al. 2007; Ting et al. 2015). Although a description of the mechanical system and its role in interlimb coordination/locomotion is outside the scope of this review, I refer the reader to other published works that have discussed this topic in mammals (Bertram 2016; Biewener 2006; Biewener and Daley 2007; Lee 2011; Lee et al. 2011; Maes and Abourachid 2013; Schamhardt 1998; Usherwood et al. 2007).

Control of forelimb-hindlimb coordination by propriospinal pathways.

Now that we have established that each limb is controlled by its own spinal locomotor CPG, how do they communicate to achieve stable yet flexible locomotion? The spinal cord is composed of a dense network of neurons with their cell body and main axonal projection contained within it (Flynn et al. 2011; Jankowska 1992). These propriospinal neurons interconnect spinal segments with axons that descend or ascend and that send projections ipsilaterally or contralaterally. Propriospinal neurons have axonal projections that can project over short (1–6 spinal segments) or long distances (>6 spinal segments) (Flynn et al. 2011; Saywell et al. 2011; Skinner et al. 1979).

The propriospinal system involved in coordinating the fore- and hindlimbs can be broadly divided into long descending and long ascending pathways. The long descending propriospinal pathways considered here are those originating in cervical/upper thoracic segments that project to the lumbosacral cord. These can be composed of propriospinal neurons with long axonal projections that directly terminate on lumbosacral neurons, or of propriospinal neurons with shorter projections that relay inputs to lumbosacral levels via interposed neurons (e.g., other propriospinal neurons or interneurons). The original anatomical studies were mainly performed in cats and monkeys and more recently in mice and rats (Brockett et al. 2013; Giovanelli Barilari and Kuypers 1969; Matsushita et al. 1979; Skinner et al. 1979; Miller et al. 1998; Reed et al. 2006, 2009; Ruder et al. 2016). These pathways have their cell bodies primarily in laminae VII and VIII, and they originate from all cervical/upper thoracic segments. They project homolaterally and diagonally to the lumbar cord, and both types of projections appear to be evenly spread in terms of synaptic terminal density (Brockett et al. 2013; Reed et al. 2006). Terminations show a rostrocaudal gradient with a greater number of terminations found in more rostral lumbar segments compared with caudal ones (Brockett et al. 2013). Electrophysiological studies have shown that some of the descending propriospinal projections have monosynaptic excitatory connections with motoneurons and also with excitatory and inhibitory interneurons that project directly to motoneurons (Jankowska et al. 1973, 1974). Pathways that are minimally disynaptic elicit stronger excitatory responses, indicating that interneuronal transmission facilitates descending excitatory propriospinal inputs. The long descending propriospinal pathways also receive dense synaptic inputs from supraspinal pathways and from neck and forelimb afferents, allowing them to integrate information related to voluntary control, posture, and head position (Alstermark et al. 1987, 2007; Brockett et al. 2013; Ruder et al. 2016).

The long ascending propriospinal pathways considered here are those originating in lumbosacral levels that project to upper thoracic/cervical segments. These can be composed of propriospinal neurons with long axonal projections that directly terminate on upper thoracic/cervical neurons, or of propriospinal neurons with shorter projections that relay inputs to other neurons that will then reach the networks controlling arm/forelimb movements. Like the descending propriospinal pathways, most of the original anatomical studies of long ascending propriospinal pathways were also performed in cats and monkeys and more recently in mice and rats (Brockett et al. 2013; Dutton et al. 2006; English et al. 1985; Giovanelli Barilari and Kuypers 1969; Matsushita and Ueyama 1973; Molenaar and Kuypers 1978; Reed et al. 2006; Sterling and Kuypers 1968). These pathways have their cell bodies primarily in laminae VII and VIII and project homolaterally and diagonally via the ventrolateral funiculus to the ventrolateral motor nuclei of lower cervical/upper thoracic segments. Pathways that project diagonally have denser synaptic terminations in the cervical cord than those projecting homolaterally (Brockett et al. 2013; English et al. 1985; Reed et al. 2006). It also appears that long ascending propriospinal pathways are primarily excitatory in rats and mice (Brockett et al. 2013; Ruder et al. 2016), although some inhibitory synaptic terminal connections, containing GABA and glycine, were found in the rat (Brockett et al. 2013). In the rat, the ascending homolateral projection is almost entirely excitatory, whereas the diagonal projection is more evenly distributed between excitatory and inhibitory (Brockett et al. 2013). The ascending pathways appear to primarily make monosynaptic contacts on motoneurons that innervate shoulder muscles, such as the pectoralis major/minor and the latissimus dorsi (Brockett et al. 2013). In the cat, it was also shown that they make synaptic contact at C₃–C₄, a region of the cervical cord containing short descending propriospinal neurons that directly project to motoneurons controlling forelimb movements (Alstermark et al. 1987, 2007; English et al. 1985).

How can long descending and ascending propriospinal pathways coordinate the fore- and hindlimbs during locomotion in mammals? This has been primarily investigated using the in vitro neonatal rat semi- or fully isolated spinal cord preparation (Cowley and Schmidt 1997; Cowley et al. 2008, 2009, 2010; Juvin et al. 2005, 2012; Zaporozhets et al. 2006). Some general organizational principles emerge from these elegant electrophysiological studies that used bath partitions, pharmacology/neurochemistry, electrical stimulation, and different types of spinal lesions. First, with the use of longitudinal sections along the midline of the spinal cord, it was shown that crossed propriospinal projections play an important role in left-right or diagonal coordination (Cowley and Schmidt 1997; Cowley et al. 2009, 2010; Juvin et al. 2012). The crossed projections are redundantly distributed along the length of the spinal cord allowing transmission to cross over to the other side at different spinal segments. Second, performing lateral hemisections on opposite sides of the thoracic cord a few segments apart, disrupting supraspinal pathways with long descending projections, did not block the initiation of hindlimb locomotor-like activity by stimulating the brainstem, indicating that signals from the brain stem can be relayed via propriospinal neurons with crossed projections (Cowley et al. 2010). Third, selectively blocking the transmission of propriospinal relay neurons or the activity within thoracic segments, without affecting the long propriospinal projections, considerably weakened cervicolumbar coordination without completely abolishing it (Juvin et al. 2012). This is consistent with a strong contribution from propriospinal relay neurons and/or distributed rhythm-generating elements within the thoracic cord in coordinating cervical and lumbar CPGs. Fourth, there appears to be asymmetric interactions between cervical and lumbar spinal CPGs, with a more powerful direct excitatory ascending influence from lumbar to cervical CPGs as well as the presence of a direct ascending inhibitory influence in the newborn rat (Juvin et al. 2005). To demonstrate this, Juvin et al. (2005) applied excitatory neuromodulators caudal to a partition that could be moved rostrally from T₁₁ to T₂. The probability that a cervical rhythm was concomitantly evoked with the lumbar rhythm increased the more rostral the partition was made in the thoracic cord, most likely by recruiting additional rhythm-generating elements and/or ascending propriospinal pathways. In contrast, applying excitatory neuromodulators rostral to the partition, up to T₁₀, which is just rostral to the hindlimb CPGs (Cazalets et al. 1995), elicited a locomotor-like rhythm at cervical levels but failed to activate lumbar CPGs. In the same study, blocking inhibitory transmission caudal to T₇ produced a slow and synchronous bilateral rhythm in both lumbar and cervical ventral roots. However, when inhibitory transmission was blocked rostral to T₇, although a synchronous bilateral rhythm was present in cervical ventral roots, the left-right alternation between lumbar (L₂) ventral roots was unaffected. These results suggest that hindlimb CPGs impose their rhythm on forelimb CPGs but not the other way around. Whether these results extend to adult rats and other species remains to be confirmed (discussed in Thibaudier and Hurteau 2012).

More recently, Ruder et al. (2016) injected the diphtheria toxin in mice whose long descending propriospinal neurons expressed the diphtheria toxin receptor. Silencing these neurons produced subtle changes during overground locomotion, such as larger angles when turning, less distance traveled, and a reduced maximal speed. At a fast treadmill speed (40 cm/s), there was a change in left-right alternation of the hindlimbs, with greater synchronous activity. Thus these data in mice suggest that long descending propriospinal neurons play a role in modulating the alternation of hindlimb activity at fast speeds or, in other words, in shaping transitions from symmetrical (e.g., trot) to asymmetrical (e.g., bound, gallop) gaits.

In summary, the spinal cord has a rich and diverse propriospinal circuitry that can coordinate cervical and lumbar spinal locomotor CPGs through long or short descending and ascending axonal projections. The thoracic propriospinal system seems to act like a conduit, changing the flow between networks controlling the fore- and hindlimbs: decreasing it to allow for independent use of the forelimbs or increasing it to strengthen forelimb-hindlimb coordination. The propriospinal system also likely plays an important role in mediating gait transitions.

Control of forelimb-hindlimb coordination by somatosensory feedback from the limbs.

The CNS is informed of the position of the body and its interactions with the environment through sophisticated sensors located in joints, muscles, and skin. These afferent inputs from mechanoreceptors exert powerful effects on the locomotor pattern by regulating phase transitions, reinforcing ongoing activity and by correcting movement trajectory in response to perturbations (Duysens et al. 2000; Frigon and Rossignol 2006; Pearson 2004, 2008; Rossignol et al. 2006; Zehr and Stein 1999). Inputs from arm/forelimb and leg/hindlimb afferents can also activate pathways that project to the spinal networks that control the limbs of the other girdle.

Lloyd and McIntyre (1948) described the actions of long descending propriospinal pathways on hindlimb motoneurons by stimulating forelimb afferents and recording the electroneurography (ENG) of lumbar ventral roots and/or hindlimb muscle nerves in high spinal (complete spinal transection at upper cervical levels) cats. These long descending propriospinal pathways projected both homolaterally and diagonally, with both excitatory and inhibitory actions on hindlimb ventral roots/muscle nerves. Schomburg and colleagues (Schomburg et al. 1977, 1978, 1986) later confirmed and extended this work by characterizing the intraneuronal effects of electrically stimulating forelimb muscle and cutaneous afferents in hindlimb motoneurons of high spinal (complete spinal transection at C₁ level) cats at rest and during fictive locomotion. At rest, forelimb afferent inputs produced a mix of excitatory (EPSPs) and inhibitory postsynaptic potentials (IPSPs) in hindlimb motoneurons (Schomburg et al. 1978, 1986). The most common responses were short-latency EPSPs, with a minimal latency of ~5 ms, followed by a long-lasting hyperpolarization, although pure EPSPs or IPSPs could also be evoked. In general, the effects were stronger in hindlimb extensors than flexors, confirming the results of Lloyd and McIntyre (1948), suggesting that these pathways could serve to reinforce extensor activity during the stance phase. The long descending propriospinal pathways could be activated by both homolateral and diagonal forelimb afferents from mixed and cutaneous nerves. Although the evoked response patterns from homolateral and diagonal inputs were similar, the effects were generally weaker with diagonal inputs. A short-latency IPSP, with a minimal latency of ~3 ms, was also observed in flexor hallucis longus (FHL) or flexor digitorum longus (FDL) with stimulation of distal cutaneous and mixed muscle nerves from the homolateral side only. The majority of responses in hindlimb motoneurons were evoked by muscle and cutaneous afferents at group II strength and higher with the exception of the IPSP in FHL/FDL motoneurons, which was consistently evoked by stimuli at group I strength (see also Fleshman et al. 1984). When forelimb afferents were stimulated in high spinal cats during pharmacologically evoked fictive locomotion, EPSPs in hindlimb motoneurons were only observed in extensors and flexors during their period of activity (i.e., extension for extensors and flexion for flexors) (Schomburg et al. 1977). Although weaker and less frequent, IPSPs could also be observed in hindlimb motoneurons in their inactive phases (i.e., flexion for extensors and extension for flexors), suggesting that the rhythmic alternation between excitation and inhibition in hindlimb motoneurons during locomotion could be provided, or reinforced, in part by descending propriospinal pathways activated by forelimb afferent inputs. The EPSPs and IPSPs observed during fictive locomotion could be evoked by both homolateral and diagonal pathways. The receptive fields of long descending propriospinal pathways were shown to vary widely in size, with some cells responding to mechanical pressure on one digit, whereas others responded to mechanical pressure applied over the entire forelimb, as studied in high spinal cats (Skinner et al. 1980). In the same study, long descending propriospinal neurons could have ipsilateral or contralateral receptive fields only, whereas some also received inputs from both forelimbs concurrently. Moreover, some cells were unimodal, responding preferentially to one type of input (e.g., cutaneous or joint movement), whereas others were multimodal.

To assess the activation of long ascending propriospinal pathways by hindlimb afferents, Miller and colleagues stimulated hindlimb muscle and cutaneous nerves and recorded the ENG of forelimb muscle nerves in high spinal (complete transection at C₁) cats (Miller et al. 1973). They also investigated the effects of stimulating hindlimb afferents on local segmental reflexes of forelimb muscles. The most consistent responses were observed in pectoralis major and minor, muscles that play a critical role in supporting and stabilizing the upper body in quadrupeds. Homolateral hindlimb afferents produced stronger responses, with a shorter latency (by ~2 ms), than diagonal ones. The most effective hindlimb afferent inputs came from the sartorius (hip flexor/knee extensor/knee flexor) and quadriceps (knee extensor/hip flexor) muscle nerves and from the sural and superficial peroneal cutaneous nerves. Stimulation intensity of hindlimb muscle and cutaneous nerves at group II strength and higher was generally required to evoke responses in motor nerves of the pectoral girdle. By lesioning the thoracic spinal cord, Miller et al. (1973) showed that ascending propriospinal pathways activated by hindlimb afferent inputs were located in the ventrolateral funiculus.

Another method of assessing neural connectivity from forelimb afferents to hindlimb motor pools, and vice versa, is to electrically stimulate peripheral nerves and record the responses evoked in muscles of the limbs of the other girdle with electromyography. In the present review, I term these interlimb reflexes, whereas reflexes between homologous limbs are referred to as crossed reflexes. The advantage of interlimb reflexes is that they can be evoked and recorded during real locomotion. Miller et al. (1977) investigated the modulation of descending and ascending interlimb reflexes in adult decerebrate cats. They showed that stimulating forelimb and hindlimb nerves evoked short-latency (13 and 18 ms in hindlimb and forelimb muscles, respectively) excitatory or inhibitory responses in homolateral muscles. Responses were modulated with phase, being mostly present during the period of activity of the muscle or slightly preceding it. Interlimb reflexes were still present after a complete transection at upper cervical levels following treatment with nialamide and l-3,4-dihydroxyphenylalanine (l-DOPA), although they were considerably weaker, suggesting that these responses are mediated by propriospinal pathways with a possible supraspinal contribution (see Gernandt and Shimamura 1961).

In summary, somatosensory feedback from the limbs activates parallel excitatory and inhibitory descending and ascending propriospinal pathways that project homolaterally and diagonally to the networks controlling the limbs of the other girdle. A general organizational principle, at least in electrophysiological studies, is that homolateral pathways, both descending and ascending, seem to be stronger than diagonal ones. Moreover, descending and ascending propriospinal pathways primarily require somatosensory inputs from high threshold afferents (group II and higher). Although some descending and ascending propriospinal pathways appear to be monosynaptic to distant motoneuronal pools, the majority are minimally disynaptic, indicating that communication is relayed through more than one propriospinal neuron. The task- and phase-dependent modulation of interlimb reflexes and the pattern of muscle-specific responses suggest an important role of these pathways for forelimb-hindlimb coordination during mammalian locomotion.

Control of forelimb-hindlimb coordination by supraspinal structures.

Areas of the brain are critical for planning and selecting movements and integrating visual information. Supraspinal structures are also essential for postural control and are informed of the biomechanical state of the body and limbs through various ascending spinal pathways. The brain and brain stem project to the spinal cord via four major descending pathways: corticospinal, rubrospinal, reticulospinal, and vestibulospinal tracts. These descending tracts can directly or indirectly influence the activity of spinal cervical and lumbar locomotor CPGs or of the pathways involved in their coordination. However, only a few studies have specifically addressed the supraspinal control of forelimb-hindlimb coordination during mammalian locomotion. In this section, the role of descending pathways from the reticular formation in controlling forelimb-hindlimb coordination is first described, because it has received more attention than other pathways, followed by studies that have shown the effects of stimulating other supraspinal structures.

Since the discovery that electrical stimulation of an area of the midbrain, the mesencephalic locomotor region (MLR), could elicit locomotion in the decerebrate cat (Shik et al. 1966) by projecting to the reticular formation, the reticulospinal tract has been a major focus in the study of locomotor control (recently reviewed in Jordan et al. 2008; Le Ray et al. 2011; Ryczko and Dubuc 2013). Reticulospinal neurons involved in the control of locomotion originate in the pontine (PRF) and medullary reticular formations (MRF) and project ipsilaterally and contralaterally to both cervical and lumbar enlargements, branching extensively throughout the mammalian spinal cord (Hayes and Rustioni 1981; Holstege et al. 1979; Liang et al. 2015, 2016; Martin et al. 1981; Matsuyama et al. 2004; Peterson et al. 1975; Peterson 1979; Sivertsen et al. 2014, 2016; Takakusaki et al. 2016). Stimulating reticulospinal pathways evokes mono-, di-, or polysynaptic EPSPs or IPSPs in forelimb and hindlimb motoneurons (Grillner et al. 1968; Jankowska et al. 1968; Peterson et al. 1979; Riddle et al. 2009). The ipsilaterally projecting pathway from the PRF appears to provide a stronger and more direct route to motoneurons than the crossed pathway (Sivertsen et al. 2014).

The effects of electrically stimulating the MRF on forelimb and hindlimb activity have been studied during treadmill locomotion in intact (Drew 1991) and decerebrate cats (Drew and Rossignol 1984), as well as during spontaneous or MLR-evoked fictive locomotion in decerebrate cats (Perreault et al. 1994). Overall, these studies show that 1) short trains of stimuli to the MRF can elicit responses in flexors and extensors of all four limbs concurrently, with about a 2- to 3-ms delay in those of the hindlimbs relative to the forelimbs. 2) MRF-evoked responses are modulated with phase, being generally maximal during their period of activity. 3) Ipsilateral flexor and contralateral extensor muscles in the limbs of the same girdle are generally reciprocally coupled. For example, stimuli to the MRF during the ipsilateral forelimb swing phase evoked excitatory responses at similar latencies in the ipsilateral forelimb flexor and in the contralateral forelimb extensor. 4) Short-latency responses to forelimb and hindlimb flexors, ipsilateral and contralateral to MRF stimulation, are generally excitatory. On the other hand, short-latency responses to forelimb and hindlimb extensors are more variable and can be inhibitory or excitatory. 5) Generally, stimulating the MRF during the ipsilateral forelimb or hindlimb swing/flexion phase prolonged the ongoing phase and flexor burst duration as well as the contralateral stance/extension phase and extensor burst duration. The same stimuli applied during ipsilateral stance/extension phase produced weaker effects during treadmill locomotion (Drew and Rossignol 1984). 6) During fictive locomotion, a common effect of stimulating the MRF during ipsilateral forelimb extension was a resetting of the rhythm, reducing cycle duration and advancing the next flexion phase (Perreault et al. 1994). The same stimuli applied during ipsilateral forelimb or hindlimb flexion considerably prolonged cycle duration along with the duration of the ipsilateral flexor and contralateral extensor bursts. Longer stimulation trains were required to obtain similar results during treadmill locomotion in decerebrate cats (Drew and Rossignol 1984). These results suggest that MRF inputs have direct access to rhythm-generating circuits of both the forelimb and hindlimb CPGs.

Microstimulation or recording studies of other supraspinal structures have also been done in the cat during locomotion. For instance, stimulating the forelimb or hindlimb representation of the motor cortex, which gives rise to the main portion of the corticospinal tract, in intact cats evoked phase-dependent responses during treadmill locomotion in forelimb and hindlimb muscles, respectively (Bretzner and Drew 2005; Rho et al. 1999). The main response pattern consisted of facilitation of flexor activity during swing and weaker facilitation or suppression of extensor activity during stance. Although these responses were prominent in contralateral muscles, some weaker responses in ipsilateral muscles were also observed. Stimulation of the forelimb or hindlimb representation could also evoke responses in muscles of the hindlimb and forelimb, respectively. For example, stimulating the hindlimb representation could prolong the activity of the contralateral cleidobrachialis (elbow flexor/shoulder protactor) during swing and the contralateral triceps brachii (elbow extensor) during stance. These motor cortical inputs also appear to have direct access to rhythm-generating circuitry, because longer stimulation trains prolonged flexion of the contralateral limb during swing, whereas during stance there could be a reset to flexion.

Stimulation of the red nucleus, which gives rise to the rubrospinal tract, evoked phase-dependent responses in forelimb and hindlimb muscles during treadmill locomotion in intact cats, which were qualitatively similar to those observed with stimulation of the motor cortex (Rho et al. 1999). Stimulating loci that primarily gave rise to responses in forelimb muscles also elicited responses in hindlimb muscles at longer latencies, particularly in the contralateral semitendinosus (knee flexor/hip extensor). One notable difference with stimulation of the motor cortex was that long trains of stimuli applied to the red nucleus failed to reset the cycle in intact cats during treadmill locomotion. Rho et al. (1999) proposed that although both pathways can modify the structure of the step cycle, the motor cortex has a more direct access to spinal neurons that control timing.

The vestibulospinal tract has been implicated in regulating the activity of extensor muscles in the four limbs during locomotion (Matsuyama and Drew 2000a, 2000b). In these studies, vestibulospinal neurons from the dorsal part of the lateral vestibular nucleus were recorded during treadmill locomotion in intact cats. Vestibulospinal neurons discharged intensely during locomotion and most frequently exhibited two peaks of activity during the step cycle. It appeared that one peak corresponded to hindlimb activity, whereas the other was related to the forelimb. A portion of these neurons had wide receptive fields, discharging in response to passive manipulations of all four limbs, suggesting that they could be involved in regulating muscle activity of all four limbs concurrently.

In summary, descending inputs from supraspinal structures regulate the coordination and/or timing of muscle activity in all four limbs concurrently. Some of these pathways can reset the forelimb and/or hindlimb locomotor pattern, indicating direct access to spinal rhythm-generating circuitry. The specific contributions of these descending pathways and how they interact with spinal circuits and somatosensory feedback to regulate forelimb-hindlimb coordination in various locomotor tasks remains, however, largely unknown.

Schematic representation of the neural control of interlimb coordination.

Figure 1 summarizes the different neural structures and pathways within the spinal cord that can play a role in forelimb-hindlimb coordination during locomotion in mammals. Each limb is controlled by its own spinal locomotor CPG (see Distinct central pattern generators for the forelimbs and hindlimbs and Distinct central pattern generators for the left and right sides), represented by a yin-yang symbol to illustrate the reciprocal interactions between parts of the CPG controlling flexor and extensor activity. Left-right interactions between spinal locomotor CPGs at cervical or lumbar levels are mediated by commissural interneurons whose axons cross the midline (Butt et al. 2002; Kiehn 2006, 2016; Shevtsova et al. 2016). Neural communication between cervical and lumbar CPGs is achieved by ascending and descending propriospinal pathways that project homolaterally and diagonally (see Control of forelimb-hindlimb coordination by propriospinal pathways). These spinal circuits and pathways are regulated by peripheral somatosensory feedback (see Control of forelimb-hindlimb coordination by somatosensory feedback from the limbs) and supraspinal inputs (see Control of forelimb-hindlimb coordination by supraspinal structures). The arrows represent direct or indirect connections that can be excitatory or inhibitory. As discussed in the following section, these neural mechanisms can be disrupted in varying degrees by spinal lesions.

Fig. 1.

Schematic representation of the neural control of interlimb coordination. Each limb is controlled by its own spinal locomotor central pattern generator (CPG). Left-right coordination at cervical and lumbar levels is mediated by commissural interneuronal pathways. Descending and ascending propriospinal pathways, with homolateral and diagonal projections, coordinate cervical and lumbar CPGs. The propriospinal pathways can be composed of neurons with long or short axonal projections. Supraspinal inputs and somatosensory feedback from the limbs access the spinal CPGs via commissural and propriospinal pathways. The arrows can represent an excitatory or inhibitory influence. E, extensor; F, flexor; LF, left forelimb; LH, left hindlimb; RF, right forelimb; RH, right hindlimb.

Control of Forelimb-Hindlimb Coordination After Spinal Lesions

Experimental studies of spinal lesions in animal models have shed some light on the neural structures and pathways potentially involved in forelimb-hindlimb coordination during real locomotion and also on the capacity for functional recovery after SCI. These lesions disrupt some of the neural control mechanisms described in the previous section. For instance, a spinal lesion at mid-thoracic level will not only disrupt supraspinal pathways, it will also disrupt descending and ascending propriospinal pathways, as well as ascending somatosensory pathways. Thus lesion studies as a means to assess the role of structures and neural pathways in the control of locomotion should be interpreted with caution because the lesions are rarely confined to specific tracts and substantial morphological and functional plasticity within remaining circuits can occur. For instance, a partial spinal lesion increases the control of hindlimb locomotion by mechanisms caudal to the SCI, such as changes within the lumbar CPG and/or in how somatosensory feedback interacts with spinal circuits (Barrière et al. 2008, 2010; Cowley et al. 2015; Frigon et al. 2009; Gossard et al. 2015). As such, the control mechanisms observed after SCI might not accurately reflect the ones present before the lesion. Nevertheless, these studies provide insight into the neural control of forelimb-hindlimb coordination during locomotion.

The work by Miller and colleagues is often cited as evidence that spinal circuits interacting with somatosensory feedback are sufficient to produce stable interlimb coordination in an adult mammal (Miller and Van der Meché 1976; Miller et al. 1977). In these studies, quadrupedal locomotion was described following spinal transection at C₁ and treatment with nialamide and l-DOPA in adult decerebrate cats. Although Miller and colleagues concluded that the spinal cord had the required circuitry to coordinate the fore- and hindlimbs, this is debatable on the basis of several observations. First, when the four limbs were initially placed on a motorized treadmill, only the hindlimbs started stepping. The forelimbs remained retracted caudally and only started stepping with the hindlimbs later on. Moreover, the forelimbs were not well lifted during the swing phase, and they could not support body weight. The forelimbs could also step in phase or out of phase with their homolateral hindlimbs, or the locomotor rhythm could be irregular. Unfortunately, no statistical analysis was made to determine the strength or consistency of forelimb-hindlimb coordination. The observation that high spinal cats can perform stepping of the fore- and hindlimbs simultaneously does not provide conclusive evidence that spinal mechanisms interacting with somatosensory feedback are sufficient for an effective forelimb-hindlimb coordination, only that forelimb and hindlimb CPGs can be concurrently active.

Various types of incomplete spinal lesions have been made to better understand the pathways involved in the control of forelimb-hindlimb coordination and/or hindlimb locomotion. For instance, English (1980, 1985) studied interlimb coordination during overground locomotion before and after bilateral lesions of the dorsal columns at mid-thoracic (T₆), low thoracic (T₁₂), and upper cervical (C₂–C₃) levels in cats (English 1980; English 1985). The bilateral destruction of the dorsal columns at T₆, T₁₂, or C₂–C₃ had no effect on left-right coordination of the forelimbs or hindlimbs. Before the lesions, cats had a predominant trotting form of locomotion, with the homolateral limbs performing an out-of-phase coupling. However, after the T₆ and T₁₂ lesions, there was a considerable shift in homolateral phasing toward a pacing-like gait (e.g., synchronous activity of homolateral limbs), despite similar stepping speeds and the maintenance of a 1:1 forelimb-hindlimb coordination (i.e., equal cycle duration and number of steps for the fore- and hindlimbs). Interestingly, bilateral destruction of the dorsal columns at C₂–C₃ did not alter homolateral coordination. This suggests that the dorsal column medial lemniscal pathway, which carries proprioceptive and exteroceptive inputs from the limbs to the somatosensory cortex, does not play a major role in forelimb-hindlimb coordination, because it was also disrupted with the cervical lesion. English (1985) also excluded the dorsal spinocerebellar tract, because lesioning this tract bilaterally had no effect on forelimb-hindlimb coordination, and also a role for secondary dorsal column neurons, because the bilateral dorsal column lesion at T₆ had similar effects on forelimb-hindlimb coordination as the one performed at T₁₂. Thus it was argued that the most likely candidate modifying forelimb-hindlimb coordination following destruction of the dorsal columns at mid- or lower thoracic levels was a long ascending propriospinal pathway.

Although the aforementioned changes in forelimb-hindlimb coordination were subtle (e.g., phase shift in homolateral coordination), a striking change after partial spinal lesions in mammalian quadrupeds occurs when the forelimb rhythm becomes dissociated from the hindlimb one. In other words, the forelimbs and hindlimbs start stepping at different frequencies. When this occurs, it is always the forelimbs that adopt a faster cadence, with two forelimb cycles taken within one hindlimb cycle. Such a dissociation between the fore- and hindlimb rhythms has been reported during treadmill or overground locomotion in rats and cats following various types of incomplete spinal lesions at thoracic levels, including lateral hemisections (Barrière et al. 2010; Thibaudier et al. 2017), ventral/ventrolateral lesions (Bem et al. 1995; Brustein and Rossignol 1998), dorsal/dorsolateral lesions (Górska et al. 1996, 2013; Jiang and Drew 1996), and contusion/compression injuries (Alluin et al. 2011; Kloos et al. 2005). Therefore, a dissociation between the fore- and hindlimb rhythms following various types of partial spinal lesions is probably not due to the loss of any one pathway (supraspinal, propriospinal, and/or ascending somatosensory) but to lesion extent. In the rat, it was shown that forelimb-hindlimb coordination was maintained if more than 25% of the spinal white matter was spared in the transverse plane, particularly in the ventral and lateral funiculi where the reticulospinal tract is located (Górska et al. 2013; Kloos et al. 2005). Although hindlimb stepping in rats can recover with as little as 10% of the spinal white matter remaining in the transverse plane, forelimb-hindlimb coordination is lost (Kloos et al. 2005; Górska et al. 2013). Górska et al. (2013) also noted that forelimb-hindlimb coordination recovered better in rats compared with cats (Górska et al. 1996) following similar partial spinal lesions, suggesting species-dependent differences.

With a 2:1 forelimb-hindlimb relationship, cycle duration and stride lengths of the forelimbs are decreased, whereas those of the hindlimbs are unchanged or increased relative to prelesion levels (Górska et al. 2013; Thibaudier et al. 2017). It should be noted that cycles with a 2:1 forelimb-hindlimb relationship are most frequently interspersed with 1:1 relationships. In widely used scales to rate locomotor performance during overground stepping after SCI, such as the Basso, Beattie and Bresnahan (BBB) 21-point scale, forelimb-hindlimb coordination is an important criterion (Basso et al. 1995). However, the BBB scale is concerned with a single aspect of forelimb-hindlimb coordination: a 1:1 step ratio. As such, the appearance of a 2:1 forelimb-hindlimb relationship is generally taken as a loss of forelimb-hindlimb coordination, and the maximal score that can be attained on the BBB scale is 10 or 11 (i.e., no forelimb-hindlimb coordination). However, only a few studies that have reported a dissociation of the fore- and hindlimb rhythms have analyzed homolateral or diagonal phasing on a step-by-step basis, considering each forelimb step separately within a hindlimb cycle. When we applied this analysis during treadmill locomotion in cats after a lateral spinal hemisection, which disrupts some of the pathways coordinating the fore- and hindlimbs (Fig. 2A), we found that homolateral or diagonal phasing was maintained when the first and second forelimb steps were treated separately (Thibaudier et al. 2017). The homolateral and diagonal phase values for the first and second forelimb steps clustered around two values on circular plots. Moreover, transverse split-belt locomotion with the hindlimbs stepping faster than the forelimbs restored a 1:1 forelimb-hindlimb relationship and increased the step-by-step consistency of forelimb-hindlimb coordination. Interestingly, if the forelimbs stepped faster than the hindlimbs, consistent 2:1 forelimb-hindlimb relationships occurred with an increased step-by-step consistency in homolateral and diagonal phasing. These results suggest that neural communication between cervical and lumbar levels is still present after a lateral spinal hemisection despite the presence of a 2:1 forelimb-hindlimb relationship and that it can be modulated by manipulating the stepping speeds of the limbs of the two girdles independently. However, we cannot exclude that this new form of coordination is mediated or assisted by mechanical linkages between the limbs and trunk. Either way, it appears that transverse split-belt locomotion generates an environment that challenges the system to adopt a more consistent gait pattern, which could have translational applications. A more consistent forelimb-hindlimb coordination could be due to a strengthening of propriospinal pathways via somatosensory feedback from the forelimbs and hindlimbs moving at different speeds and/or to enhanced supraspinal drive. Whether such observations apply to other types of spinal lesions remains to be investigated.

Fig. 2.

Schematic representation of the effects of various spinal lesions on the neural control of forelimb-hindlimb coordination. Shown are the effects of a mid-thoracic lateral hemisection (A), contralaterally placed lateral hemisections at mid-thoracic and low thoracic levels (B), and contralaterally placed lateral hemisections at upper cervical and mid-thoracic levels (C) on the neural control of interlimb coordination depicted in Fig. 1. The dashed lines for diagonal pathways indicate a weakening of these projections with the various lesions. CPG, central pattern generator; E, extensor; F, flexor; LF, left forelimb; LH, left hindlimb; RF, right forelimb; RH, right hindlimb.

The reason for this compensatory strategy (i.e., the appearance of 2:1 forelimb-hindlimb coordination) remains unclear. Górska et al. (2013) suggested that a faster forelimb cadence could be due to a reduced inhibitory influence from hindlimb CPGs to forelimb CPGs or to a shift in the center of gravity rostrally so that the forelimbs bear a greater proportion of bodyweight after SCI. The fact that intact cats also produce a 2:1 forelimb-hindlimb relationship when the forelimbs step faster than the hindlimbs on a transverse split-belt treadmill suggests that it is probably not the result of impaired communication between forelimb and hindlimb CPGs (Thibaudier et al. 2013; Thibaudier and Frigon 2014). Instead, Thibaudier et al. (2017) proposed that a 2:1 forelimb-hindlimb relationship was a way to maximize dynamic stability. For instance, smaller forelimb steps could prevent the forelimbs from interfering with the hindlimbs at the forelimb stance-to-swing transition and the hindlimb swing-to-stance transition. Moreover, performing smaller forelimb steps could be a strategy to maintain the center of gravity within the support polygon, which is the surface obtained by joining the different points of contact of the animal on a surface (Cartmill et al. 2002).

Another type of spinal lesion paradigm that has been used to study forelimb-hindlimb coordination and/or hindlimb locomotor control consists of performing lateral hemisections placed on opposite sides of the spinal cord at different levels, either simultaneously or following an interval of time (Courtine et al. 2008; Cowley et al. 2015; Kato et al. 1984; Stelzner and Cullen 1991; van den Brand et al. 2012). When placed at thoracic levels, these lesions bilaterally disrupt the long supraspinal pathways that project to the lumbar cord and propriospinal pathways with long axonal homolateral projections (Fig. 2B). They also disrupt some of the axonal projections from propriospinal neurons that project over a few segments and those that project diagonally. However, some propriospinal pathways remain to communicate between cervical and lumbar levels, and it was shown that the remaining propriospinal circuitry has the capacity to transmit inputs from the brain stem to the lumbar cord (Cowley et al. 2008, 2010). Following contralaterally placed lateral hemisections at thoracic levels, quadrupedal locomotion recovered spontaneously (i.e., without drugs or training) in rats and cats, although this was facilitated when balance and/or weight support was provided (Cowley et al. 2015; Kato et al. 1984; Stelzner and Cullen 1991). However, forelimb-hindlimb coordination (i.e., consistent phasing between homolateral or diagonal limbs) was permanently lost during treadmill locomotion or overground locomotion (Cowley et al. 2015; Kato et al. 1984; Stelzner and Cullen 1991). Neurochemical excitation of the propriospinal circuitry between the rostral and caudal thoracic hemisections improved quadrupedal locomotion but failed to restore forelimb-hindlimb coordination (Cowley et al. 2015).

Kato et al. (1984) also performed a lateral spinal hemisection at C₂ that was followed several weeks later by a second lateral hemisection at T₆ contralaterally. Despite sparing much of the propriospinal pathways on one side (Fig. 2C), forelimb-hindlimb coordination was also lost. Moreover, cats also lost the ability to transition to trot and gallop following both types of contralateral hemisections (cervical-thoracic and thoracic-thoracic), gait patterns that were present in the intact state and following a single hemisection (Kato et al. 1984). Instead, cats could only perform a walking type of gait pattern. This inability to perform gait transitions could be a deficit resulting from the loss of specific pathways or to a general loss of stability. Thus, with contralaterally placed lateral hemisections, it is likely that the hindlimbs are being driven primarily by spinal mechanisms and somatosensory feedback below the caudal hemisection (Barrière et al. 2008; Frigon et al. 2009; Gossard et al. 2015; Stelzner and Cullen 1991). Even though some supraspinal inputs can still reach lumbar levels via propriospinal neurons, they are insufficient to restore forelimb-hindlimb coordination. Altogether, these results suggest that the propriospinal relay system is insufficient, by itself, to fully support forelimb-hindlimb coordination in adult mammals. However, as stated, spinal lesions induce anatomical and functional changes within the spinal cord and the control mechanisms observed after the lesion might not be an accurate representation of those normally present in an intact system.

In summary, incomplete spinal lesions in experimental mammalian models can alter or completely abolish forelimb-hindlimb coordination. The degree of impairment in forelimb-hindlimb coordination is dependent on the extent of the lesion, and it does not appear that any one pathway is essential. Forelimb-hindlimb coordination appears to be a higher level function compared with hindlimb locomotion, because its control requires more spared structures/pathways to recover. Although forelimb-hindlimb coordination is already one of the main outcome measures of locomotor performance scales, more sensitive measures to quantify it, such as the step-by-step phasing of homolateral or diagonal limbs, need to be incorporated because new strategies can emerge to coordinate the fore- and hindlimbs, possibly to maximize dynamic stability.

Control of Arm-Leg Coordination During Locomotion in Humans

The previous sections described the neural mechanisms involved in forelimb-hindlimb coordination in animal models. An important question is, how similar are humans compared with other quadrupedal mammals? This is important because an easy and often repeated critique of biomedical research in quadrupeds is the apparent lack of transferability to adult humans that walk on two legs. It is important to remember that human locomotion is quadrupedal during the first locomotor stage of life. Patterns of interlimb coordination observed in infants crawling are similar to those in other quadrupeds, such as dogs and cats, consistent with a conservation of basic neural circuits controlling locomotor behaviors in mammals (Patrick et al. 2009, 2012; Righetti et al. 2015). Other forms of locomotion in human adults that use the arms for propulsion, such as crawling and different forms of swimming, also involve coordinated activity of the arms and legs (Wannier et al. 2001). Although bipedal walking in humans certainly has particular demands and undoubtedly some specialized neural control mechanisms, it has been suggested that a neural control similar to that in other quadrupeds has been conserved (Dietz 2002, 2011; Dietz and Michel 2009; Dominici et al. 2011; Ivanenko et al. 2013; Tan 2006; Zehr and Duysens 2004; Zehr et al. 2009, 2016). Indeed, although movement of the arms is dispensable to maintain dynamic stability during bipedal walking, at least in undemanding conditions, the arms remain rhythmically coordinated with the legs. It has been shown that rhythmic movement of the arms during bipedal walking is generated not only by passive biomechanical linkages but also by neural commands (Ballesteros et al. 1965; Craik et al. 1976) and that these are most likely generated by spinal locomotor CPGs controlling rhythmic arm and leg movements (Dietz 2002, 2011; Dietz and Michel 2009; Zehr and Duysens 2004; Zehr et al. 2004, 2009, 2016). The anatomical presence of propriospinal pathways has also been identified in humans (Nathan et al. 1996). That the human neural control systems somehow dismissed millions of years of evolution when making the transition from quadrupedal to bipedal terrestrial locomotion would be surprisingly wasteful. Instead, the human CNS most likely integrated new control mechanisms into circuits already present to meet the need for new functional demands.

Interlimb reflexes coupling the arms and legs in humans.

Interlimb reflexes, from the arms to the legs or from the legs to the arms, have also been identified in humans in static conditions (Calancie 1991; Calancie et al. 1996, 2002, 2005; Delwaide and Crenna 1984; Kearney and Chan 1979, 1981; Meinck and Piesiur-Strehlow 1981; Zehr et al. 2001) and during locomotion (Dietz et al. 2001; Haridas and Zehr 2003; Haridas et al. 2006). Zehr et al. (2001) reported that the shortest latency responses were observed at ~30–40 ms in arm and leg muscles with stimulation of the superficial peroneal (SP) and superficial radial (SR) nerves, respectively. These short-latency responses are consistent with transmission through propriospinal pathways. Interestingly, some interlimb reflexes were only evoked in homolateral or diagonal limbs. For instance, responses in the anterior deltoid were only observed with diagonal SP stimulation, whereas responses in the posterior deltoid were only evoked with homolateral stimulation. Similarly, short-latency responses with SR stimulation were consistently evoked in the homolateral soleus but not in the diagonal one. Interlimb reflexes from the legs to the arms increased in amplitude from standing/sitting to walking (Dietz et al. 2001). These reflexes were also modulated with phase during walking (Dietz et al. 2001; Haridas and Zehr 2003). Interestingly, during walking, short-latency responses (≤75 ms) in arm and leg muscles with SP and SR nerve stimulations, respectively, were less frequently evoked, whereas middle-latency (80–120 ms) responses were prominent (Haridas and Zehr 2003). This contrasts with short-latency (≤75 ms) interlimb reflexes evoked during isometric contractions (Zehr et al. 2001) and could suggest that pathways involved in arm-leg coordination during walking in humans are strongly modulated by supraspinal pathways and/or by movement-related phasic somatosensory feedback. It was later shown that interlimb reflexes from the foot to arm muscles were strengthened when walking in an unstable environment, but only when the arms were held statically in a crossed position (Haridas et al. 2006). When the arms moved rhythmically, interlimb reflexes were reduced when postural threat increased. Thus task-dependent modulation of interlimb reflexes during walking is also dependent on the state of the arms (e.g., rhythmic movement vs. static position).

In summary, the arms and legs in humans appear to be coordinated by neural pathways that are activated by somatosensory feedback. The short-latency component of these interlimb reflexes are consistent with the activation of propriospinal pathways, whereas the modulation of the longer latency component suggests a supraspinal contribution. Although the specific contributions of these pathways to human locomotion remain largely unknown, they could be involved in strengthening arm-leg coordination in tasks requiring greater dynamic stability.

Arm-leg coordination following an incomplete spinal cord injury in humans.

Similar to quadrupedal mammals, injury at any level of the spinal cord will disrupt neuronal circuits and/or pathways involved in coordinating the arms and legs in humans. Not surprisingly, the coordination between the arms and legs is often altered after SCI, or even absent (Visintin and Barbeau 1994; Tester et al. 2011, 2012). Tester et al. (2011) reported that the majority of their participants with incomplete SCI (American Spinal Injury Association Impairment Scale grade C or D) did not perform arm swing during walking (18 of 30 subjects), indicating impaired communication between the spinal networks controlling arm and leg rhythmic movements. Overall, individuals that did not use assistive devices in their daily lives (e.g., cane, crutches) were more likely to display arm movements and had better walking performances. Of the 21 participants that received 9 wk of locomotor training, 16 did not initially perform arm swing during walking. However, following the training, 8 of 16 participants integrated arm swing during walking, suggesting that locomotor training can re-engage the circuits coordinating arm and leg activity.

In a follow-up study, Tester et al. (2012) compared arm-leg coordination in healthy subjects and in people with incomplete SCI over a range of treadmill speeds. Although both groups showed coordinated arm-leg phasing during treadmill walking, there were some notable differences. First, healthy human subjects showed a coordinated 2:1 arm-leg coupling (2 arm cycles for 1 leg cycle) at slow walking speeds (≤0.5 m/s) and a 1:1 coupling at comfortable walking speeds and higher (≥0.8 m/s), as shown previously (Donker et al. 2001). On the other hand, subjects with an incomplete SCI used a 1:1 arm-leg coupling at all speeds tested, even at the slowest speeds tested. Second, the phasing between movements at the shoulder and contralateral hip joint (i.e., diagonal coupling) became more consistent with increased walking speed in healthy subjects, whereas in people with incomplete SCI, the strength of the diagonal coupling was already maximal at slow walking speeds so that little modulation was observed with increasing speed. This suggests that people with incomplete SCI make use of a more consistent coupling pattern between the arms and legs during walking, particularly at slow speeds, possibly to maximize residual function and dynamic stability. However, as stated by the authors, the lack of modulation with increasing speed, due to the strong coupling between the arms and legs at slow speeds, might reflect a less flexible and more stereotypical movement pattern.

Several authors have suggested that incorporating natural arm movements could promote the recovery of walking in people with SCI by engaging circuits that communicate between cervical and lumbar levels (Behrman and Harkema 2000; Dietz 2011; Dietz and Michel 2009; de Kam et al. 2013; Ferris et al. 2006; Klimstra et al. 2009; Shah et al. 2013; Thibaudier et al. 2017; Zehr et al. 2009, 2016). In line with this, one study showed that leg muscle activity induced by passive locomotor-like movements was modulated by passive and active arm movements in subjects with an incomplete SCI at cervical levels (Kawashima et al. 2008). Interestingly, voluntarily moving the arms in rhythm with the legs generated greater modulation of leg muscle activity in some subjects compared with passive movements. Visintin and Barbeau (1994) also reported that movement of the arms improved the pattern of leg muscle activation in subjects with incomplete SCI. Altogether, these results suggest that somatosensory feedback from the moving arms can activate the spinal networks controlling leg activity in people with incomplete SCI, most likely via long descending propriospinal pathways, and that supraspinal inputs can reinforce this activation.

In summary, the coordination between the arms and legs is often impaired, or even absent, in people with incomplete SCI. Locomotor training can facilitate the recovery of arm movements during walking. Thus, similar to forelimb-hindlimb coordination in preclinical mammalian models, arm-leg coordination appears to be an important outcome measure of locomotor performance. Incorporating normal arm swing into locomotor training or daily life could be an effective and natural approach to strengthen descending pathways that project to spinal networks controlling leg movements during walking.

Concluding Remarks

In this review, the flexibility and complexity of the neural structures/pathways potentially involved in coordinating the spinal locomotor CPGs controlling rhythmic movements of the arms/forelimbs and legs/hindlimbs were discussed. Evidence was presented that each limb is controlled by its own spinal locomotor CPG and that these networks can be coordinated by descending and ascending propriospinal pathways. I also discussed the role of somatosensory feedback in activating propriospinal pathways and the potential role played by interlimb reflexes during locomotion. Different supraspinal pathways also undoubtedly contribute to specific aspects of interlimb coordination, because stimulating pathways from the reticular formation, the motor cortex, and the red nucleus can elicit responses in all four limbs simultaneously. How these different mechanisms are integrated to produce stable interlimb coordination remains largely unknown and will remain an active area of research.

Although substantial progress has been made and new tools are now available, major gaps in knowledge remain in identifying the specific role played by different structures/pathways in the neural control of arm/forelimb and leg/hindlimb coordination. The genetic manipulation of specific neuronal populations and/or the selective ablation of specific neurons and/or pathways using toxins constitutes a powerful approach to dissect the neural organization of locomotor networks (Bellardita and Kiehn 2015; Ruder et al. 2016; Talpalar et al. 2013). However, the CNS is a closed-looped system, and silencing specific neuronal populations most likely leads to immediate functional and compensatory changes in other circuits and pathways. Thus these approaches are not without their own limitations. Moreover, we cannot lose sight that a primary goal of biomedical research is to translate these results to real-world applications, such as motor rehabilitation for people with various sensorimotor disorders. In this respect, neural control mechanisms and strategies might differ substantially between small rodents and larger mammalian species that have a higher center of gravity and limb segments with greater mass and inertial properties. As stated earlier, how neural mechanisms interact with properties of the musculoskeletal system remains poorly understood, even more so during complex multilimb movements.

Another limitation of most experimental studies is that one structure or pathway is often studied in isolation to determine its specific contribution in a restricted behavioral context (e.g., self-selected overground locomotion or treadmill locomotion at a single speed). How postural control is integrated into the control of interlimb coordination and locomotion also remains largely unknown. Ultimately, in vivo mammalian preparations where physiological processes and interactions at multiple levels of the nervous system can be studied concurrently along with intrinsic properties of the musculoskeletal system over a wide range of locomotor behaviors, from simple to complex, will be important in elucidating the neural and biomechanical mechanisms controlling interlimb coordination. To achieve this goal, computational models of neural networks controlling the four limbs (Buono and Golubitsky 2001; Collins and Richmond 1994; Danner et al. 2016; Maufroy et al. 2008), derived from biological data obtained in rich and complex behavioral contexts, that are tested and refined by controlling biomechanical models (Dzeladini et al. 2014; Markin et al. 2016) and/or quadrupedal or human-like robots (Aoi et al. 2013; Fujiki et al. 2015; Hunt et al. 2015; Pfeifer et al. 2007; Spröwitz et al. 2014) will be critical in providing a more complete understanding of the neuromechanical control of interlimb coordination. This new knowledge will be important in developing novel strategies that aim to tap into spared neuronal circuits and pathways to promote motor recovery in people with SCI and other movement disorders that affect the coordination between the limbs.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author.

AUTHOR CONTRIBUTIONS

A.F. interpreted results of experiments; A.F. prepared figures; A.F. drafted manuscript; A.F. edited and revised manuscript; A.F. approved final version of manuscript.