Spinal automaticity of movement control and its role in recovering function after spinal injury

ABSTRACT

Introduction

The significance of the spinal cord in controlling postural and locomotor functions largely reemerged in the mid-1970s under the leadership of Sten Grillner, demonstrating key phenomena of ‘central pattern generation’ and ‘fictive locomotion’ with an evolutionary perspective. These concepts raised the question of how much function can be recovered after paralysis, given the intrinsic automaticity of spinal networks in injured and uninjured states in adults.

Areas covered

This review explores biological mechanisms governing spinal control of movements such as posture and locomotion. We focus on concepts that have evolved from experiments performed over the past decade. Rather than a comprehensive review of the vast literature on the neural control of posture and locomotion, we focus on the various mechanisms underlying functional automaticity, and their clinical relevance.

Expert opinion

We propose that multiple combinations of sensory mechanoreceptors linked to proprioception that generate an infinite number of different sensory ensembles, having species-specific meaning and extensive influence in controlling posture and locomotion. These sensory ensembles are translated as a probabilistic phenomenon into highly specific but indeterminate actions. Therefore, we opine that spinal translation of these ensembles in real-time plays a central role in the automaticity of motor control in individuals with and without severe neuromotor dysfunction.

KEYWORDS:

• Spinal control
• locomotion
• feedforward
• automaticity
• interneurons

1. Introduction

One of the most complex tasks for the nervous system is to control movements while simultaneously updating multimodal inputs to all the motor pools driving the actions. The nervous system accomplishes this feat by changing spinal networks to physiological states congruent with the action intended. The physiological state is defined largely by the composite of all sensory modalities. Consequently, the arduous task of generating an infinite array of potential movements requires the extensive transformation of sensory inputs to motor actions in real-time. These highly integrated actions, for the most part, must occur ‘automatically’ or even ‘autonomically.’ Furthermore, for real-time automaticity to occur as effectively and smoothly as possible, the spinal and supraspinal sensory networks must function in a feedforward and synergistic manner. In this review, we present evidence for the dominating role of spinal networks in controlling movement. While the premise is based on mammalian experiments, it is obvious that the major concepts are variations of invertebrate systems. Also, while much of this review is focused on the control of movement in uninjured states, most concepts are derived from earlier studies on spinal injuries. Stuart and Hultborn [1] have elegantly summarized the history of motor control physiology with a focus on spinal networks at the systems level, while Kiehn [2] has presented a more of a circuit and cellular perspective.

A Prologue. It is almost impossible to imagine that when the Society for Neuroscience was formed in 1969 that the pervasive view of the nervous system was that it was virtually incapable of significant levels of plasticity in adult mammals, particularly in humans. This transformation in thinking has evolved slowly, carrying the message: be cautious in accepting dogma. The degree to which the focus has been extensively migrated ‘to the tree and away from the forest’ is also remarkable. As greater detail becomes known about the mechanisms of controlling behavior, it has become more evident that the control of movement is derived from mechanisms of different levels of integration of multiple sensory modalities across multiple organ systems. The relative importance of the multidimensional and integrated functions in controlling the behavior is highly dynamic, as reflected in the constantly changing physiological states of the spinal as well as supraspinal and neuroendocrine networks. New technologies are providing the opportunities to explore the gap between a single neural cell and a behavior of an organism in in vivo conditions, and thus allowing quantitative assessments of multiple controlling mechanisms at multiple levels and kinds of integrative events. In a summary of the book of a major symposium, The Interneuron, published in 1967, it was bluntly stated that an interneuron was ‘ignorant.’ That is, it cannot be aware of which combinations of neurons is contributing to its physiological state at any given time, nor can it be aware of the other sources of input that the neurons to which it projects is receiving at any given time. From this perspective, it is clear that any action is the result of a highly integrated and probabilistic responses. In identifying mechanisms of the control of movement, a necessary assumption must be that, how any isolated component performs in a reduced experimental preparation, is likely to be markedly different in its normal highly integrated environment.

We will begin by focusing on the anatomical and functional properties of the better-known components of the motor system, recruitment of motor units of a single motor pool.

2. Two basic biological phenomena among spinal networks play a primary and essential role in controlling movement

The first phenomenon has been studied extensively and deservedly is referred to as a “principle “and is commonly referred to as such, the ‘size principle,’ which characterizes the rank order of recruitment of motor neurons within an individual motor pool being controlled by a series of parameters related in multiple ways to the anatomical size of multiple features of the motor unit, i.e. the motor neurons as well as the muscle fibers innervated by that motor neuron. It was largely characterized by Henneman and colleagues [3] (see Mendell, 2005 for summary [4]) and other authors to follow such as Heckman and Enoka [5]. The significance of the size principle in the control of behavioral features in vivo has been reviewed by Burke and Edgerton [6].

The second phenomenon of primary importance in the mechanisms of control of movement is related to the pattern of recruitment of different motor pools in a given motor task which defines how a movement is coordinated and in effect, determines what the movement task will be. The mechanisms underlying this control of the coordination of motor pools has also been studied extensively, but its complexity has precluded a substantial understanding of the basic mechanisms to reach a level of clarity that justifies the formulation of a principle. Both phenomena highlight an extensive degree of automaticity, which minimizes the need for constant, detailed, conscious awareness in controlling movement.

2.1. Defining which and how many motor units will be recruited within a single motor pool

According to the size principle, the order of recruitment of motor neurons within a given motor pool starts from small to large. The number of muscle fibers innervated by a single motor neuron provides an accurate estimate of the size of a motor unit and the order of recruitment within each motor pool ([7] and (Figure 1(a,b)). Furthermore, the number of motor units activated within a motor pool determines the force, speed, and power generated by the muscle. The orderly recruitment of muscle fibers of a muscle unit confers biochemical and physiological benefits. The slow oxidative motor unit can perform the most work and has the highest resistance to fatigue (Figure 1(a)). Thus, smaller units tend to be recruited more often compared to larger, more easily fatigable motor units. The size principle assumes that recruiting the slow (S) motor units with a high oxidative capacity, is a more effective strategy for generating movements requiring low force and power. In contrast, the fast fatigable (FF) units relying primarily on glycolytic metabolism generate higher forces, speed, and power, but are needed less often (Figure 1(b,c)). The fast fatigue-resistant (FR) units are more intermediate in force and fatigability.

Figure 1. (a) A conceptual illustration of the primary anatomical, metabolic, and physiological properties of different types of motor units. The neural component of the motor unit consists of dendrites that make up as much as 90% of its surface area. All muscle fibers innervated by a single motor neuron are referred to as a ‘muscle unit.’ The different colors among the muscle fibers of a unit represent the unique biochemical profiles of the muscle fibers within a muscle unit. However, all of the fibers within a unit are generally homogeneous, having similar metabolic and physiological properties that will vary from fiber-to-fiber within a muscle unit, only about 5%. Motor unit phenotypes are typically designated as Fast Fatigue (FF), Fast Fatigue Resistant (FR), and Slow (S) units, reflecting properties of the motor neuron and its specific muscle unit. Muscle units are commonly designated as Fast Glycolytic (FG), Fast Oxidative Glycolytic (FOG), and Slow Oxidative (SO) [6 from modifications figure 8A,p 53]. The two fibers stained brown (ATPase activated) and tan (ATPase activity after being inhibited with a specific acid pH) signifies the fast myosin phenotype of which signifies the speed of contraction. (b) The relationship between the total cross-sectional area of all muscle fibers within a single unit relative to the total tetanic tension that can be generated by that motor unit. Note that the slow units (crosses) generate a smaller tetanic tension per total cross-sectional area of all fibers within each unit [7]; (c) The relationship between the tetanic force generated by single motor units and the cumulative force that would be generated from motor units assuming recruitment of the units with a nonlinear increase in tetanic forces among the population of motor units within the cat medial gastrocnemius [8]. When a motor pool is engaged in any movement, the rank order of recruitment for each motor pool will proceed from the smallest (fewest number of muscle fibers) to the largest motor unit i.e.highest number of muscle fibers within that motor pool.

2.2. Defining patterns of recruitment among motor pools

The second principle is not as well characterized mechanistically as the size principle. However, what the second principle functionally entails is very clear. It defines the pattern of recruitment of motor pools and the approximate number of motor units per motor pool that are activated in each phase of a given motor task. Here, we present evidence to demonstrate that the pattern of activation of all the motor pools required to perform a specific motor task is how they are coordinated. In addition, we propose that the pattern of coordination of motor pools that defines the nature of any movement is derived principally from sensory input, with proprioception being the dominant sensory mode.

We know that there are continuously changing spinal ‘interneuronal pools’ that translate the sensory signals and supraspinal descending input to activate a specific combination of motor pools (Figure 2(a,b)). The constantly changing sensory ensemble during a motor movement like stepping is derived largely from proprioception. We suggest that the sensory ensemble selects a phase-specific combination of interneurons that, in turn, will activate the next phases of a set of motor neurons to generate the unique action that will continue the stepping cycle (Figure 2(b)). The uniqueness of each phase-dependent sensory ensemble is recognized by a specific combination of interneurons that match the uniqueness of the sensory ensemble and define the pattern of motor neurons and motor pools that will be activated. Previous studies have reported that an animal with a mid-thoracic, complete spinal transection can step successfully when a treadmill belt is moving backward, forward, or at a sideward angle [9]. This finding demonstrates that cutaneous and proprioceptive information derived from the mechanics of stepping enables full weight-bearing stepping over a range of speeds and directions. Interestingly, stepping forward, backward or sideways recruits many of the same motor pools, but the patterns of activation vary for each type of movement. When proprioception recognizes the body’s gradual orientation toward the right or left, the kinematics of the limbs almost perfectly match the proprioceptive ensembles generated by gravitational vectors when walking toward the right or to the left. To summarize, unique patterns of proprioception can drive a unique combination of interneurons, which in turn drive a unique pattern of excitation or inhibition to a unique combination of motor pools (Figure 2(b)).

Figure 2. (a) Drawing of the cervical spinal cord of a cat by Santiago Ramon y Cajal [10]. This classic anatomical sketch of the input projecting from the dorsal segment of the gray matter, passing through a cluster of interneurons, demonstrates a challenging perspective as to how these interneurons within this cluster defines which excitatory and inhibitory signals functionally project to either, generally the more medial flexor or, laterally located extensor motor pools in the most ventral part of the gray matter. (b) Cartoon sketch to illustrate the hypothesis that different combinations of interneurons are activated by a specific combination of proprioceptors (green curve). Such sensory ensembles involving unique combinations of interneurons mediate a unique pattern of activation of motor pools. The pattern of motor pool activation controls the next phase of the motor output, ultimately defining the next specific phase of a planned movement. The translation of sensory signals occurs in real-time. However, the motor outcome is ‘planned’ in a feedforward manner by interneurons that contribute to central pattern generation for repetitive tasks such as locomotion.

Spinal interneurons functionally project to multiple motor neurons of multiple motor pools. The continuously changing sensory ensembles drive a predictable change in the pool of interneurons and predictable network of motor neurons for the next phase of a movement (Figure 2(b)). However, translating sensory input into a motor action requires an estimated 60–100 milliseconds. Such a delay in translating the sensory cues between successive actions would result in jerky movements similar to those observed by repetitively triggering monosynaptic and polysynaptic reflexes. Smooth, coordinated movements can be generated by sensory ensembles of proprioceptive patterns being translated into action in a predictable feedforward or appropriately matched planned event. We have previously reviewed the substantial evidence for this feedforward design of spinal neural networks [11].

3. Unique features of proprioception

An evolutionary perspective suggests a far greater role for proprioception in routinely defining millisecond-to-millisecond motor responses than seems to be generally recognized. To limit the challenges associated with understanding automaticity, we have focused on relatively immediate sensorimotor responses [12,13]

Proprioception provides rapid feedforward information about the spatial location of each body segment at any given time. Prochazka and Gorassini concluded in 1998: ‘Our overall conclusion is that to a first approximation, large muscle afferents in the cat hindlimb signal muscle velocity, muscle length, and muscle force, at least in movements of the speed and amplitude seen in locomotion’ [14]. To further emphasize the potential influence of spindle afferents in shaping spinal networks, they estimated that assuming that all the muscle afferents in a limb fire at comparable mean frequencies (around 80/sec and assuming about 10,000 of them in the hindlimb), the mean input to the spinal cord during locomotion would be 800,000 action potentials/sec in a single step cycle of a cat. Visual input may serve as a substitute for proprioception, but its response time is much too slow, routinely requiring approximately 1000 ms between the visual trigger and a kinematic response [15,16]. Instead of proprioception from the hindquarters of a cat being processed supraspinally and then instructing the spinal circuitry to excite the intended action, we propose that the fact that proprioceptive information is continuously conveyed directly to distal motor pools amplifies it importance. The significance of proprioception is apparent in individuals who have lost all proprioception of the lower body and are severely paralyzed despite having anatomically normal supraspinal inputs (including vision) to the lower body.

Proprioception can serve as the primary driver of the size principle that governs the order of recruitment of each motor pool. In addition, it generates the sensory ensembles that coordinate the activation of multiple motor pools. Forssberg et al [17] provided compelling evidence of the key role of proprioception in generating coordinated movements, and the relationship between proprioception and central pattern generation. For example, this work demonstrated so clearly that the same tripping perturbation to the dorsum of the paw during the swing phase of a step that triggered a bilateral response of the hindquarters consisting of a flexion response on the side ipsilateral to the tripped limb but an extensor response contralaterally. If however, the same perturbation was applied during the stance phase of the limb, the opposite responses were observed. Thus, the responses were context-dependent and of high functional significance. The importance of the feedforward function of proprioception in the step-to-step variability of kinematics through a series of step cycles was demonstrated by Musienko et al. [18]. A definitive example of the feed forwardness of this proprioceptive input from one step to the next is shown in Figure 3(a–c), again demonstrating that the physiological state of the spinal networks determines the actions that will be generated. The responses are not defined by only the input. The step-to-step modulation and adaptation of the bipedal stepping kinematics in a decerebrated cat demonstrates that proprioceptive input is crucial for maintaining balance of the hindquarters with minimal to no input from the brain. Balance of the hindquarters persists despite substantial variability in the sequence of foot placements (Figure 3(c)), and consequently, the center of mass.

Figure 3. Feedforward regulation of maintaining balance during stepping in decerebrated cats facilitated by epidural spinal cord stimulation. (a) The cat is secured in a stereotaxic frame. An accelerometer is placed on the pelvis to record displacements, and force sensors are placed beneath each belt to record ground reaction forces (GRF) from the right and left hind limbs. (b) Right and left GRF. (c) Correlation between left and right total GRF during stepping for 10 experiments in 7 decerebrated cats (overall r = 0.98). (d) There is no correlation when the order of the left-right lateral displacements was randomized. (e) Cumulative right and left limb displacements are plotted in order of occurrence (gray line) or randomized (red line) [18]. Note that the red line (accumulation of rendomized GRFs’) diverts from the level necessary to maintain the equilibrium of the hindquarters after about 5 steps. (Modified from [18]).

This observation is significant because it is often assumed that only the vestibular system helps maintain the balance from step to step. If this were the case, an individual with a complete spinal injury would have little to no ability to use their lower limbs for maintaining equilibrium when standing and stepping. However, cats trained to step bipedally after undergoing complete midthoracic spinal transection, can readily generate 10–20 consecutive steps without the hindquarters collapsing [18]. These data suggest that a significant level of control of equilibrium may be derived from matching proprioceptive input to lumbosacral spinal networks between consecutive left and right hindlimbs, without the involvement of descending vestibular spinal inputs. The variability in the step-to-step adjustments of the electromyography (EMG) pattern and ground reaction forces (GRF) throughout the swing-stance phases of the step cycle was accommodated effectively by the spinal neural networks. Even after complete isolation from the brainstem or forebrain, lumbosacral circuits receiving tonic spinal electrical stimulation can effectively maintain equilibrium during standing and stepping [19]. The magnitude of lateral placement of the paws on consecutive steps was highly correlated between the left and right limbs (Figure 3(c)). When the same sequence of the center of pressure and placements of the paw were randomly distributed for the order that occurred during stepping, it was statistically determined that the hindquarters would have collapsed within five steps (Figure 3(d,e)). Similar results were reported after a complete mid-thoracic spinal transection [20]. These two sources of control – supraspinal and spinal networks, provides one of many examples of the redundancies that exist in the sources of motor control.

We suggest that proprioceptive, including cutaneous, input generated by the GRFs or displacements generated within any single limb alone, cannot successfully sustain equilibrium. Instead, a close match of proprioceptive input derived from the hindquarters throughout the preceding steps helps maintain equilibrium in the subsequent step. Thus, a constantly updating feedforward process provides global details of the activation of the spinal circuitry and defines the motor command that will successfully execute the subsequent step on sustaining equilibrium of the hindquarters. This feedforward design of the spinal networks aligns closely with the feedforward design of proprioception [11] and of the spinal networks that contribute to central pattern generation [21]. A range of spinal feedforward mechanisms have been reviewed by Gerasimenko and coworkers [11].

4. Significance of variability in the control of movement

4.1. Functional redundancy of networks

One of the earliest experiments to assess variability in the control of movement in humans quantified the kinematics of multiple joints and the trajectory of the hammer of a highly skilled blacksmith [22] . Subsequently, many experiments have demonstrated a near-infinite number of kinematic variations while achieving the same endpoint. This mechanism known as the redundancy phenomenon is built into our sensorimotor system, which provides the ability to perform the same motor task with different combinations of neural networks (Figure 3(b)). The amount of variation, even for a task as fundamental as stepping, highlights the importance of this design feature in the neural control of all movements.

Early experiments recorded interneurons during fictive locomotion using a preparation in which the lumbosacral segments of the adult mammalian spinal cord were surgically separated from the brain and upper spinal segments. Denervation of peripheral nerves and pharmacological paralysis of the neuromuscular junctions in these preparations ensured that the lumbosacral segments were devoid of any oscillating sensory input from the periphery. During these experiments, thousands of cyclic bursting activities were recorded from motor nerves innervating the flexor and extensor muscles of multiple hind limb segments. Simultaneously, extracellular recordings of action potentials were obtained from interneurons throughout the dorsal and ventral gray matter of lumbosacral spinal segments. Despite the constancy of the tissue environment, there was variation in the burst-to-burst properties. The variation was reflected as changes in the amplitude of individual action potentials and the duration of the burst, as well as in the period length of the cycles of central pattern generation. These observations demonstrate that even without any extrinsic oscillatory input, variability is a fundamental and intrinsic aspect of the design of motor control [23]. This intrinsic variability becomes apparent simply by observing the infinite anatomical and physiological variations present at multiple levels of the sensorimotor system [24,25]. Therefore, the outcome from any given translation of sensory ensembles or supraspinal descending commands is a probabilistic event.

The automaticity that is intrinsic to spinal networks and the translation via the neuromuscular interface of proprioception into highly specific motor events are major facilitators of movement, posture, and locomotion. Importantly, sensorimotor networks caudal to a spinal lesion can be neuromodulated to physiological states when combined with locomotor training, enabling completely paralyzed human subjects to generate voluntary movement, including full weight-bearing stepping [26–31]. Similarly, animal experiments demonstrate that when proprioceptive signals become disorganized following an injury, the spinal networks reorganize into novel and functional sensorimotor circuits through learning and other activity-dependent mechanisms [32–34]. Figure 4(a) shows the stick diagram decomposition, trajectory of the ankle and EMG of selected hindlimb muscles while stepping of a rat with complete mid-thoracic transection of the spinal cord. The left panel shows steps taken by injured rats before they were trained, while the right panel demonstrates the variability of stepping by the same rats after they had been trained daily for several weeks [9,35]. These observations are consistent with other studies reporting multiple possible alternatives in performing the same motor task [36]. Thus, applying selective patterns of proprioception as an intervention to drive neural reorganization via activity-based mechanisms can selectively reinforce and re-connect neural networks to confer an improved functional state and level of automaticity of movement after a spinal cord injury. The variability in the kinematics and EMG patterns in the untrained state after the injury is elevated, but after training this variability is reduced noticeably, a response that is consistent with learning and improving the skill in performing a specific task. When the neural networks have been functionally reorganized, based on EMG and kinematics, stepping quality improves. This step-to-step variability as illustrated previously in Figure 3 is a fundamental feature of the control system for stepping and in fact all movements. A predominant source of the variability in all sources of input that drives the motor pools to generate a motor task is the proprioceptive input to the dorsal horn of the spinal cord. The relative activity of the spinal neurons in the gray matter is much higher in the more dorsal layers of the dorsal horn than any other area of the spinal cord (Figure 4(b)). We counted the number of neurons in the spinal cord of a mouse that were activated, based on c-fos expression, across two 30-minute bouts of stepping on two separate occasions. Only 20% of the neurons were co-labeled as being active after performing approximately 7,000 steps on each of the two treadmill bouts (Figure 4(c)) [37]. This finding indicates that different combinations of neurons can generate thousands of steps, with each one being biomechanically similar but neurally distinct. Thus, the variability in performing repetitive movements may be attributed to 1) anatomical and physiological differences among different populations of neurons, and 2) mechanisms that generate changing synaptic dynamics over timeframes ranging from milliseconds to months.

Figure 4. (a) Stick diagram decomposition of a step cycle, joint angular displacement, and EMG activity of selected hindlimb muscles during testing in a representative, Uninjured, Spinally injured nontrained and step Trained spinal rat with epidural spinal cord stimulation (40 Hz at L2) and Quipazine (0.3 mg/kg). Modified from [35]. (b) The spatial distribution of the relative density of activated neurons in a cross-section of the lumbosacral spinal cord. Note the markedly greater density of active cells in the most dorsal lamina compared with the medial or ventral lamina, demonstrating the continuous input of sensory information from a TRAP mouse while resting. (c) Targeted recombination in active populations (TRAP) of mice allowed the capture of two different c-fos activation patterns in the same animal. The percentage of activated neurons that were co-labeled (c-fos+ and Td Tomato+) after performing two 30-minute bouts of stepping were quantified. About 7000 steps were taken in each bout. These data are consistent with a probabilistic-like phenomenon that can recruit many combinations of neural populations (synapses) when repetitively generating many step cycles. Only about 20% of the neurons activated from the first bout of stepping were also activated by the second bout (adapted from [37].

It seems obvious that a fundamental design feature of our neural control of movement imposes some level of probability as opposed to a deterministic outcome. Some critical level of variability is essential for repetitive movements such as walking. There are ‘families’ of kinematic and neural solutions that can solve the problem of placing the foot within a space that is sufficient for maintaining the necessary equilibrium with an acceptable probability for success [38]. An alternative perspective on variability would be to view the redundancy as ‘abundance’ since it provides access to multiple solutions to perform the same task. This abundance in design expands the possibilities to 1) execute an almost unlimited number of combinations of movements for a given species, 2) generate a range of adaptive responses to perturbations through kinetic and kinematic variations in the same task, 3) minimize synaptic fatigue and 4) reorganize physiological or anatomical networks in the event of neuromuscular injury or pathology.

The issue of variability in movements has been addressed extensively by Edelman [23], but with a focus on the anatomical and physiological features of the brain and the theoretical implications of the pervasive presence of variability in its design. Bizzi and colleagues have explored a different perspective by examining nervous system design features related to the concepts of ‘primitives’ and ‘synergisms’ [39], based on the probability of a motor response to a specific stimulation site or muscle. Also, Ivanenko and colleagues [40] have examined the probabilities of different combinations of muscle activation patterns, i.e. ‘synergies,’ during locomotion. Although each of these aspects is tightly linked to the basic phenomenon of defining coordination patterns, the mechanisms responsible for the emergence of these highly coordinated patterns in real-time remain unclear.

4.2. Activity-dependent modulation and plasticity of redundant networks

Variability in routinely performed and practiced motor tasks, such as locomotion, is a fundamental design feature of the neural control of movement. Compared with physiological states, complete spinal transection induces higher variability in neural control and the biomechanics of stepping. In the early phases of recovery, an injured animal may only succeed in performing a few steps in a sequence. Nonetheless, after several days or weeks of training, the step-to-step variability reduces and a greater number of consecutive steps can be performed successfully [41,42], demonstrating that the spinal networks learn qualitatively even in the injured state.

Since variability is a fundamental property of the control of movement, we tested whether variability is also an essential feature of recovering motor function after a complete mid-thoracic spinal cord transection. Rats were trained to step on a treadmill bipedally using a robotic arm with no variability in the kinematics (fixed trajectory of the ankle) or with limited robotic guidance (controlled flexibility in ankle trajectory). Consequently, rats trained with some flexibility learned to step more effectively than those with no variability. These data show that the position of the leg as planned by the spinal networks differed from that imposed by the robot. In effect, the neural networks were being constantly corrected, resulting in extensive co-contractions of the flexor and extensor muscles of the lower limb, thus disrupting effective learning ([20,43] (Figure 5(a–c)). These results present a key advantage of the probabilistic design feature as being one of abundance. These observations also have important implications in programing algorithms that control robotic devices for rehabilitation purposes. Physiologically, this reflects the fact that all aspects of our movements are probabilistic rather than deterministic.

Figure 5. (a) Ankle trajectory of a rat robotic arm during stepping in the assist as needed (AAN) paradigm. The designed trajectory is indicated in blue and the window within which stepping occurs is in red. The black trace represents an example of the pattern of the trajectory and the corrective forces applied by the robotic arm to maintain the ankle within a trajectory of a selected width, (b) Representative EMG from the ankle flexor (green) and extensor (red) during stepping under the influence of a fixed trajectory versus an AAN trajectory. (c) Average EMG of a normalized step cycle shows higher levels of co-contraction during stepping between antagonistic muscles in the fixed versus AAN mode. Modified from [43].

5. The biology of automaticity of sensory-motor function after spinal cord injury (SCI)

5.1. Built-in automaticity and synergism in the design of the functional and anatomical interface of spinal networks and muscle properties

A key element of well-designed neuromodulatory interventions is the ability to facilitate the intrinsic automaticity of spinal networks that survive an injury, such that they develop a functionally synergistic interface with supraspinal networks. As discussed earlier, the size principle governs the automaticity of posture and locomotion in the uninjured and injured states by defining the order and level of recruitment of motor unit phenotypes within each motor pool [4,6]. This design feature automatically activates motor neurons that project onto the muscle fibers best equipped to meet the mechanical and metabolic demands of the motor tasks being performed. Thus, low force-power movements predominantly recruit slow motor units and slow oxidative fibers that are designed to be fatigue resistant. It is important to note that almost all of our daily activities are accomplished using a small number of these fatigue-resistant motor units within a single motor pool. In contrast, muscle fibers that are used less often are more fatigable (Figure 1).

Following SCI, muscle recruitment may not adequately match the functional demands of a task. For example, some motor unit conversion to the faster types occurs in paralyzed individuals due, at least in part, to chronically low levels of post-injury activity. Proper use of spinal electrical neuromodulation under such circumstances can activate interneuronal networks that functionally project to populations of motor neurons and muscle unit phenotypes that meet the work demands [44].

After SCI, two factors related to force generation become relevant. If the objective is to recover muscle mass from the rapid and severe atrophy that ensues following SCI, direct muscle stimulation may be used for elevating force generation potential. The size principle plays a primary role in controlling force generation because it defines the number and size of the units per motor pool to be activated and dictates the automatic events involved in this process. In general, the force potential of a muscle is directly associated with the physiological cross section of the muscle. However, whether direct muscle stimulation gives rise to disruptive patterns of activation of sensory axons remains unknown. Nonetheless, both muscle atrophy and phenotype conversion events can be countered through well-structured activity-dependent interventions [45]. The second consideration is the quality of the movement, which is defined by the pattern of coordination of motor pools and muscles. This challenge requires a transformative reorganization of the residual spinal networks along the entire spinal axis, both rostral and caudal to the lesion and within the segments that have been injured. In addition, the spinal networks must reorganize into more functionally synergistic states to regain supraspinal sensorimotor control. Abundant evidence suggests that use-dependent training mechanisms, combined with neuromodulation, can reorganize spinal networks to achieve functional coordination among muscles following complete spinal injury [46]. However, the interactions of multiple mechanisms involved in regaining the ability to recruit and generate coordinated patterns after complete spinal injury remain undefined.

Effective strategies for restoring supraspinal control have been previously demonstrated. van den Brand [47] trained rats to step and climb bipedally to retrieve food after eliminating all long descending input from the brain to the lumbosacral spinal cord. They found that step training alone did not engage cortical neurons in stepping. But, when rats were trained to step to retrieve food, it engaged cortical neurons that could activate the lumbosacral segments and enabled the rats to retrieve a food reward. This result suggests that a voluntary task may be recovered when propriospinal neurons relay information that triggers the stepping circuitry below the thoracic lesion. By encouraging active participation with food rewards, the training paradigm triggered a cortex-dependent recovery.

Urban et al. [48] explored an auditory strategy for muscle function recovery after SCI wherein all long descending corticospinal projections were eliminated as used by van den Brand [47]. We tested whether a highly specific audio signal used to trigger a learned conditioned lower limb flexor response in rats before an injury could be relearned after a complete loss of supraspinally derived descending input caudal to a mid-thoracic lesion. A qualitatively unique audio-specific response was recovered after two months of daily lumbosacral epidural tonic neuromodulation combined with the presentation of the audio signal during spinal neuromodulation. The motor response was triggered only when the specific audio tonic epidural stimulation was presented. This finding strongly suggests that a completely novel, functionally specific supraspinal-spinal network of neurons was formed. It also indicates that sufficient guidance mechanisms were present along the novel supraspinal-spinal axis to make this new connectivity functionally relevant. Additional data demonstrated that this novel connectivity was independent of the networks that restored stepping a month before the learned audio response. Finally, it should be noted that this highly predictable and specific newly formed audio-to-motor response occurred only in the presence of a tonic, sub-motor threshold modulation. This, again, serves as an example of the automaticity that is intrinsic to a rather complex transformation of the novel supraspinal-spinal network(s).

Another example of the automaticity that is fundamental to the control of movement is the translation of highly dynamic and complex ensembles of proprioception to highly predictable motor actions. Proprioceptive inputs not only contribute to the automaticity of movement but also provide a mechanism for immediate translation of proprioceptive ensembles into action, without being burdened by ‘decision making’ about which units should be activated. We suggest that this translation is an automatic and passive event, i.e. not requiring any ‘processing’ or ‘decision-making’ related to the recruitment of motor units. However, repair mechanisms involved in the recovery of a motor response requiring the coordination of multiple motor pools are decidedly less understood than the automaticity linked to the size principle in activating a single motor pool.

Given the combination of what we propose to be key anatomical and functional features of the neural control of movement, how can such a system maneuver such a wide array of routine or complex movements with the necessary and expected precision? We suggest that proprioception plays a central role in modulating the spinal and supraspinal networks involved in the performance of such a wide array of motor tasks. Proprioceptive sensors, in conjunction with other sensory modalities, provide the spinal networks with a continuous update on the position and kinematics of all body segments. The spinal neuronal network then translates the incoming proprioceptive ensembles to the appropriate motor pools to generate the intended action. Presumably, the proprioceptive input is predictable within a reasonable probability. Since spinal networks are responsive and adaptable to use-dependent mechanisms, the certainty and precision of translation from proprioception to spinally defined actions are amenable to improvement. Feed forwardness is a key factor in the ability to effectively translate proprioception to action. As stated previously, feed forwardness is pervasive in the synaptic interactions through which proprioception, in conjunction with the net assessment of all sensory modalities from supraspinal networks, can be translated into a specific action. Another key assumption is that the proprioceptive ensembles have acquired conceptual meaning through phylogenetic, ontogenetic, and epigenetic time frames, allowing the translation of proprioceptive and other sensory modalities to generate the intended action. Given that this generalization of fundamental importance, a detailed explanation of the evidence strongly supporting it is needed and deserving of a focused effort.

5.2. Elements of adaptive synergisms among organ systems

From a physiological perspective, autonomically controlled organ systems, such as cardiovascular, bladder, bowel, sexual organs, and sweat glands, are likely to be more robustly controlled by spinal networks and ganglia than sensorimotor organs that are primarily controlled more directly by supraspinal networks [49]. Thus, the dysfunctional states of autonomic organ systems following SCI seem less formidable than the challenge of recovering sensorimotor function. We reason that because, at some early developmental stage, neural control mechanisms of the bowels, bladder, cardiovascular flow, temperature regulation, and respiration were highly effective, perhaps some level of this independence from supraspinal voluntary control may persist in the adult after spinal injury. Particularly intriguing are the organ systems controlling bowel and bladder because a new source of control normally emerges a few years after birth in humans. The ability to voluntarily control bladder and bowel functions perhaps emerged phylogenetically with evolutionary forces that may have been linked, at least, teleologically to the social demands of humans. For other mammals, however, such voluntary control seems to be less essential but learned. Can we then reengage the automaticity of bladder function, which was once present at the neonatal stage, in adults who have suffered SCI? Specifically, can neuromodulation and training be used to control bladder function and regain the sense of bladder fullness, the ability to void voluntarily, increase bladder capacity, reduce incontinence and urgency, and reduce the incidence of bladder infection [50–53] Similar questions are relevant for regaining bowel function by enhancing the ease and reducing the time required for bowel evacuation.

Some recovery of sexual function after SCI also has been reported, so that pharmacological means were no longer necessary to achieve a penile erection [25,51]. Temperature sensation in the trunk region improved in some subjects following SCI. This includes enhanced blood circulation to more rapidly warm the legs when they have been cooled inadvertently. In some subjects, neuromodulation and neuromotor training accelerate blood flow to the legs. Sweating has reappeared with training and the heart rate is elevated to more normal levels in response to more active exercise states [54]. Subjects who are chronically and severely hypotonic when attempting to sit upright or supported in a standing position without transcutaneous neuromodulation can maintain a more normal blood flow to the brain with a single session of neuromodulation of select spinal networks [55]. Similar improvements have been observed in the ability to breathe and cough independently [55,56]. In general, if the impaired organ systems are under autonomic control, the likelihood of regaining significant function seems rather positive as transcutaneous neuromodulation procedures are developed further. These autonomically controlled organ systems reflect a higher level of automaticity compared to organs under sensory-motor neural control. In addition, chronic spinal neuromodulation in patients induces neuroplasticity that improves bladder function even in the absence of electrical stimulation and can sustain the improved function for several months without neuromodulation therapy [50].

6. Conclusion

It is almost impossible to realize that when the Society for Neuroscience was formed in 1969 that the pervasive view of the nervous system was that it was virtually incapable of significant levels of plasticity in adult mammals, particularly in humans, but is now recognized by many of the degree to which the whole organism can adapt sufficiently to maintain homeostasis under a wide range of environmental and pathological challenges. This multidimensional integration of functions in controlling behaviors is highly dynamic, as reflected in the constantly changing physiological states of the spinal as well as supraspinal and neuroendocrine networks. New technologies are providing the opportunities to explore the gap between a single neural cell and behavioral tasks. In a summary of the book of a major symposium, The Interneuron, published in 1967, it was bluntly stated that an interneuron was ‘ignorant.’ It is unaware of which combinations of neurons are contributing to its physiological state at any given time, nor can it be aware of the other sources of input to which it projects at any given time. From this perspective, any action is the result of a highly integrated and probabilistic response. In identifying mechanisms of the control of movement, a necessary assumption must be that, how any isolated component performs in a reduced experimental preparation, is likely to be markedly different in its normal highly integrated environment. Thus, the relative importance of any specific control mechanism is highly dependent on the specific conditions at which it is being defined and the physiological states of the multiple networks to which it receives and projects.

The relative importance of different sensory systems in shaping the pattern of output of the sensorimotor cortex and what the spinal networks receive will be largely dependent on the specific nature of the sensations being projected. The dynamics and complexity involved in generating predictable behaviors only seems feasible when there are specific design features which automatically translates the meaning of any sensory ensemble, as opposed to there being some active ‘decision process’ occurring among neural networks. The potential for this translation of extremely complex, constantly changing sensory ensembles, only seem conceivable with a design that has evolved via hundreds of millions of years of ‘evolutionary learning,’ that is specific for any given specie. We propose that all multimodal sensory ensembles are translated with a critical level of probability in completing the ‘intended’ action. This seems to be most evident among spinal networks’ ability to translate proprioceptively-derived ensembles to specific actions. There is a high probability that at least some of these spinal networks are those that contribute to central pattern generation [57]. Then, the challenge in understanding some of the key components of the nervous system that serves as sources of control is to identify how the supraspinal derived ‘decisions’ in performing a given task. We suggest that a varying multitude of spinal networks and the massive amount of continuously arriving proprioceptive ensembles, function synergistically with a level of automaticity similar to that of other sensory modalities. We routinely translate a sensory modality, but more often combinations of sensory modalities, which forms perceptions of different qualities of odor, taste, sound, or light that generates a wider range of perception that can be translated toward a probable action. The major challenge in understanding how normal behavior is controlled requires an understanding of how multiple sensory modalities become integrated in a way that enables the organism to generate an infinite number of behaviors with a critical level of predictability.

7. Expert opinion

Proprioception is arguably the most important source of sensory information in controlling movement and becomes even more critical after motor dysfunctions. Proprioception provides instantaneous and continuous updates of the position of all muscles, joints, and body segments in space. Ultimately, the integration of all these components shapes one’s behavioral responses. Inputs from other sensory receptors, such as vision, odor, and taste provide additional assessments of one’s environment and physiological states, including happiness, sadness, and pleasure. These physiological states are visually evident in the manner in which a person is walking, dancing, the expressions on their face, and their posture.

Most complex behaviors are produced by the integration of multiple sensors signaling the physiological states of both the brain and the spinal cord. Except for new technologies in recording from the brain, movement is the only outlet for the brain to communicate. Notably, the translation of sensory inputs to actions is mediated by multiple sensory modalities that are continuously changing, providing information about the kinetics and kinematics of massive amounts of tissues to the rest of the nervous system in real-time.

Proprioception is crucial for normal control of movement. The proprioceptive state, at any point in time, provides a feedforward signal to the brain when it is forming an intent to initiate a given motor task. In other words, proprioception is used by the brain to plan a motor response and to inform the brain of a mistake. It places the motor pool that drives each muscle in an activated or inhibited state of excitability to execute the intended movement. This property of feed forwardness explains why a person who has lost all proprioception of the body is functionally paralyzed until they learn with great difficulty and extensive determination to use vision as the primary source of information about the spatial orientation of their body parts. The brain cannot plan a movement if it does not know where the body parts are located. Simply put, a GPS cannot chart a route to your destination if it does not know from where you are starting. It appears that vision can be used to guide some movements, but cannot provide the planning information needed with the immediacy necessary to move normally.

Proprioceptive input is available continuously, updating interneurons and motor neurons about where neuromuscular tendinous tissues have moved from, and where they most likely are programmed to go next. In other words, it is better than real-time. The translation of sensation-to-action occurs as a predictor of where to move next, i.e. it is continuously planning ahead. Due to its planning design or feedforward role, proprioception is critical in performing routine movements such as stepping, standing, and balancing. However, there seems to be limited awareness of the importance and robustness of this feedforward design feature. Instead, proprioception has historically been recognized more often for its role in providing ‘feedback’ to correct a movement. But what is typically assumed to be ‘feedback,’ is a rapidly planned, largely stereotypical, feedforward response.

Perhaps, the oversimplified interpretation of proprioception from standard textbooks leaves one with the impression that the only notable sensory processing that occurs in the spinal cord is what is routinely called a ‘reflex.’ In most cases, a reflex is described as a response to an unexpected situation, e.g. anticipating the presence of a curb that you are approaching, but were unaware of, when stepping. This conveys the impression that proprioception is only important for correcting and executing an unplanned movement. From this perspective, the physiological distinction between a response being a ‘reflex’ or an automatic response is conceptually fuzzy.

Generally, the feedforwardness of movement control is attributed to sensory cues detected and processed by the brain rather than the spinal networks, as previously reviewed [8]. A drawback of the current approaches used to recover sensorimotor function from paralyzing and pathological states is the emphasis on utilizing the most advanced state-of-the-art technology with insufficient awareness of its compatibility with human biology. For example, the design strategies of some brain interfaces to recover motor tasks is to bypass the neural networks within the spinal cord by projecting signals from the brain directly to muscles. Can the state-of-the-art engineered control systems outperform the neural control system within the spinal cord? The brain and spinal networks have evolved as a single organ system over hundreds of millions of years. One might ask, are we smarter than the trial-and-error strategy of over 300 million years of evolution?

Many forms of neuromodulation are being developed to treat dysfunction of sensorimotor systems as well as those under classical autonomic control. Multiple similarities exist in the activation and responsiveness of these two systems, suggesting some common mechanisms underlying the formation of new neural connections that may be available after a spinal injury.

Another approach deserving further consideration is the use of neuromodulatory strategies to modulate the excitability of neural networks approperate to the intended action without exceeding the motor threshold. In procedures that are currently used, the functional targets of modulation of the excitability of these networks are predominantly below the motor threshold. While a sub-threshold level of excitability is not sufficient to directly activate a neuron to generate action potentials, it has multiple advantages. Modulating excitability levels provides a feedforward mechanism to plan for a wide range of movements. Such movements may be initiated and controlled by proprioception, by spinal neuronal networks that remain intact after injury, and by descending inputs via the propriospinal system or even supraspinal inputs that reach the spinal lesion area. The key advantage of having these viable sources of input is that most, either directly or indirectly, project onto the interneurons and motor neurons that generate highly coordinated movements once the level of excitation exceeds their motor thresholds.

Finally, many neuromodulatory strategies and devices continue to be developed and tested. Although this review focuses primarily on the control of movement in uninjured states, virtually all the concepts are derived from earlier studies on a wide array of spinal injuries. The general principles and hypotheses addressed above are focused on how the physiological states of neural networks can be modulated, using primarily spinal electrical neuromodulation. Pharmacological modulation, however, can also be a very effective facilitator of the recovery of locomotor functions following SCI [32,45]. Nevertheless, extensive experimentation using SCI models has demonstrated that optimal locomotor recovery may be achieved by a critically defined combination of electrical and pharmacological modulation [35,57,58,59]. A hotly contested debate in the field revolves around which neuromodulation techniques are the safest and most effective. We recommend developing multiple devices and strategies that have previously demonstrated greater than expected functional recovery with adequate safety. Different devices offer alternatives and provide the possibility of being more effective for a given dysfunction in a patient-specific manner. Novel approaches should continue to be developed and tested to expand our options to more fully take advantage of the remaining functional potential of surviving networks that could eventually reach the clinical toolbox.

Article highlights

• The functional scope and importance of proprioception in regaining sensorimotor function is commonly underestimated.

• In general, the number and order of motor neurons recruited within each motor pool and the coordination of the pattern of recruitment of motor pools defines the specific movement.

• The quality and quantity of recovery of sensorimotor functions following paralysis are highly influenced by activity-dependent mechanisms.

• These activity-dependent mechanisms provide guidance in reorganizing neuron-to-neuron connectivity following neural injury.

• A fundamental and persistent design feature of spinal and supraspinal neural networks is that it functions as feedforward systems.

• Feed-forwardness enhances the automaticity in the execution of our sensory-motor behaviors.

Declaration of interest

VR Edgerton has shareholder interest in Onward and SpineX. P Gad has shareholder interest in SpineX. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

Raeesa Gupte (freelance medical writer and editor) provided writing assistance, language editing, and proofreading services. We also would like to thank Yury Gerasimenko for the many discussions and experiments shared, from which many concepts related to the control of movement evolved. Finally, we thank Hui Zhong for her never-ending contributions in playing key roles in the success of many experiments.

Additional information

Funding

This research was funded in part by Dana & Albert R. Broccoli Charitable Foundation, Nanette, and Burt Forester, including matching by PwC LLP and Roberta Wilson, Walkabout Foundation, and HINRI.