Neurorehabilitation in the 21^st Century: New Science, New Strategies, New Expectations

Abstract

Neurorehabilitation is among the most vibrant areas of biomedical research. Its main strategy has been skill-specific practice, which often fails to produce adequate recovery. Now, new recognition of CNS plasticity, new understanding of skills, and new technologies provide new strategies that enhance the efficacy of practice. The substrate of a skill is a network of neurons and synapses that extends from cortex to spinal cord and is now called a heksor. A heksor changes continually to maintain the key features of its skill, the attributes that make the skill satisfactory. Muscle activity and kinematics may change; key features are maintained. Heksors share neurons and synapses. Through their concurrent changes, they keep the CNS in a negotiated equilibrium that enables each to maintain its skill. When CNS damage occurs, the goal is to enable damaged heksors to repair themselves. Two new strategies enhance the efficacy of skill-specific practice. One increases plasticity. A damaged heksor shapes the additional plasticity through practice. The other targets beneficial plasticity to a critical site in a damaged heksor. This improves practice, enabling the heksor to achieve wider beneficial plasticity. In animals and humans, protocols that combine these strategies with practice enhance lasting recovery. The challenge is to develop, optimize, and validate these combined protocols. Computational modeling can accelerate the process. Controlled trials and comprehensive outcome assessments are essential. Pre-morbid factors and physiological measures may identify biomarkers that can predict efficacy or guide patient-specific protocol design. Many combined protocols will be noninvasive and suitable for home use.

Introduction

The advances of the past half-century have transformed the hardwired CNS of 1970, which could change in only a few places and only in limited ways, into the highly plastic CNS of today, in which change is ubiquitous. It occurs everywhere, has many mechanisms, and continues through life. Neurons change and new neurons may appear, dendrites and axons change, synapses change, glia change, hormonal influences change, even vasculature changes.^1–19 While this new recognition of lifelong CNS plasticity affects every area of neuroscience, its implications for neurorehabilitation are most remarkable.

In 1970, neurorehabilitation was a primarily clinical field; research, especially basic research, was limited and had modest goals. In 2025, neurorehabilitation is rapidly becoming one of the most exciting areas of biomedical research. With realization that the CNS remains plastic through life, therapeutic objectives thought impossible only several decades ago are now realistic; even complete recovery of function is conceivable. New understanding of how skilled behaviors are acquired and maintained generates new strategies for maximizing recovery of skills impaired by CNS injury or disease. These new strategies can use new biological, bioengineering, and behavioral technologies that extend their reach and enhance their efficacy. Our goal in this article is to accelerate translation of the new therapeutic possibilities into therapeutic realities.

In pursuit of this goal, the paper first describes neurorehabilitation as it was in the late 20^th century, summarizes the key scientific and technological advances of the past 30 years, and explains how these advances clarify the primary purpose of neurorehabilitation and provide new therapeutic strategies. It explicates the new strategies and reviews a number of new therapies based on them that are now in various stages of development. It emphasizes the value of protocols that combine the new strategies with skill-specific practice, discusses the steps necessary for their successful clinical translation, and focuses on several that target impairments of locomotion and hand/arm function that affect many millions of people. Most of these protocols are noninvasive. They have been effective in initial human studies and are now beginning translation into clinical use. Through their clinical impact, they should encourage and guide development of comparable protocols that target other sensorimotor impairments and, as understanding and practice advance, cognitive and behavioral impairments as well.

The Past and the Present

Neurorehabilitation in the 20^th Century

Neurorehabilitation was initially defined as a complex medical process intended to produce recovery from a nervous system injury, and to minimize and/or compensate for any associated functional deficits.^21–22 While the earliest reference to neurorehabilitation as a specialty area of medicine was a 1994 paper by the prominent neurologist, Fletcher H. McDowell,²¹ the new field was clearly evident some decades earlier. This is exemplified by the Northwestern University Special Therapeutic Exercise Project (NUSTEP) for physical therapy educators.²³ The 1967 NUSTEP textbook presented the field as it existed up to the 1990’s, when the Decade of the Brain (1990–2000) began to emphasize the importance of motor learning and adaptation in neurorehabilitation.^24–26 This new emphasis presaged remarkable progress in understanding both the basic science and the clinical practice of this relatively new medical specialty.

The 20^th-century neurorehabilitation described in the NUSTEP textbook consisted mainly of the neurofacilitation approaches of the prevailing gurus of the time (e.g., Bobath, Brunnstrom, Knott and Voss, and Rood). Their rationale was rooted in the existing knowledge of neurophysiology, motor development, motor learning, and motor behavior. They assumed that it was not possible to restore CNS areas that had been destroyed, but that it was possible to facilitate desirable movements with appropriate physical handling or positioning of the patient. This strategy was based on animal and human studies of the complex reflexes that are evident early in life and are subsequently suppressed during development (e.g., asymmetric tonic neck reflexes²⁷). These primitive behaviors may remain in people with developmental disorders (e.g., cerebral palsy), or they may reappear later in life after a spinal cord injury (SCI), a stroke, or other CNS damage (e.g.,²⁸). The prevailing idea was that loss of normal descending inhibitory control from higher centers (e.g., cerebral cortex) allowed lower centers in the brainstem or spinal cord to produce these primitive behaviors, such as abnormal muscle synergies.²⁹ Thus, therapists tried to suppress these behaviors or use them to substitute for higher-level behaviors that had been lost.

For example, the Bobath approach used reflex-inhibiting postures to diminish spasticity or abnormal tone.³⁰ Movements looked better, but this improvement required the ongoing participation of a skilled therapist; the patient alone could not produce these improved movements. Thus, these temporary changes in motor behavior did not translate to improved motor function outside the therapy session. Basically, these approaches were more about what a skilled clinician could elicit through proper handling and tactile stimulation, and less about how therapy could promote recovery of voluntary goal-directed actions or skills. Because there was no evidence that anything the clinician could prescribe (e.g., massive amounts of skill-specific practice) could induce lasting beneficial change in the brain or spinal cord (i.e., neuroplasticity), therapists simply sought to minimize spasticity and abnormal posturing, and to thereby enable use of whatever residual capacity remained.

In the atmosphere created by these limited expectations; neurorehabilitation differed widely from one center to the next. Some clinicians taught caregivers and patients to perform functional activities, such as getting up from a chair and walking, by using specific actions to minimize hypertonicity and abnormal posturing (e.g., holding clasped hands in front while getting up from a chair). Others thought that compensatory behaviors or assistive devices exaggerated abnormal muscle synergies and spasticity, and thus discouraged their use. Furthermore, even though it was apparent that muscle weakness often contributed to impaired movement, many clinicians ascribed to the now discredited idea that strengthening exercises exaggerated spasticity, and thus did not prescribe them.^31–33 This was the field in the early 1990’s.

The Transition to Now

In the 30 years since then, scientific and technical advances have revolutionized the science of neurorehabilitation, and have thereby introduced the potential for similarly revolutionizing the practice of neurorehabilitation. New recognition of lifelong CNS plasticity (e.g.,³⁴), coupled with new understanding of motor learning ushered in an era in which challenging skill-specific practice combined with new methods for inducing and guiding beneficial plasticity sought to restore lost function.^25,35–38 For example, by the beginning of the new millennium the Bobath approach to stroke rehabilitation had evolved in theory, terminology, and methods. The latent capacity for neuroplasticity became its primary rationale; therapists sought to use afferent input to target beneficial change in the damaged CNS.³⁸

Newly energized research associations such as the World Federation of Neurorehabilitation (WFNR), the European Federation of NeuroRehabilitation Societies (ECNR), the American Congress of Rehabilitation Medicine (ACRM), and the American Society of Neurorehabilitation (ASNR) began to promote energetic exchange of ideas and encourage growth of basic science and clinical publications. This transformation was exemplified and accelerated in the United States by the National Center for Medical Rehabilitation Research (NCMRR) of the National Institutes of Health (NIH). Established in 1990, NCMRR increased funding for rehabilitation research and developed a comprehensive plan for the field.³⁹ The World Health Organization (WHO) developed an inclusive description of disabilities that extended the therapeutic mission of neurorehabilitation beyond addressing functional impairments to addressing also the impact of these impairments on a person’s activities and participation in society. Its International Classification of Functioning, Disability and Health (ICF) defined the three ICF domains of Body structure/function (impairments), Activities (limitations), and Participation (restrictions).²⁴⁰ More recently, the International Stroke Recovery and Rehabilitation Roundtable (SRRR) Taskforce standardized terminology and assessment methods.^40,41 Particularly striking evidence of the transformation driven by the new recognition of life-long CNS plasticity is the prominence of neurorehabilitation in the scientific literature. Of the 148 peer-reviewed rehabilitation journals, three of the top five focus on neurorehabilitation.⁴²

This new focus also reflects greater societal and programmatic recognition and appreciation of the problems faced by people with severe neuromuscular disabilities, the important contributions these individuals can make when their problems are adequately addressed, and the value of having them participate in the development and implementation of new therapies.⁴³ This attention is very timely, for increased life expectancy coupled with the improved survival of people with neurological disorders such as stroke, traumatic brain injury, SCI, and Parkinson’s disease has greatly increased the worldwide prevalence of people who can benefit from neurorehabilitation.⁴⁴ Disorders that were usually fatal in the past are now becoming chronic and often highly disabling diseases. Thus, creation and dissemination of the better therapies made possible by recent scientific and technological advances grows more and more urgent.

The new neurorehabilitation therapies now possible fall into two classes, long-term (i.e., on-going) and short-term (i.e., transient). Effective long-term therapies do not cure a disorder such as SCI, stroke, or Parkinson’s disease; they mitigate its functional effects. If the therapy stops, its benefits disappear. Thus, these therapies become a permanent part of a person’s life. Perhaps the most prominent current example is deep brain stimulation (DBS), which can markedly improve movement in people with Parkinson’s disease or essential tremor.^45,46 In contrast, effective short-term therapies do cure, or at least partially cure, a disorder, and thereby restore more normal function. When the therapy ends, function may remain better, and may even improve further. Both long-term and short-term therapies have much to offer; and both will doubtless become increasingly effective as understanding and technology advance. That said, the most fundamental and important implication of the scientific and technological advances discussed in the next section is that restoration of normal function is possible. Thus, this article focuses on short-term therapies that produce persistent functional improvement. And it focuses primarily on noninvasive therapies for disorders of locomotion and hand/arm function because several of these therapies are ready for clinical translation. Through their success, they can be harbingers and drivers of comparable therapies for a wide range of sensorimotor, cognitive, and behavioral impairments.(e.g., ^{241, 242})

Scientific Advances

Fifty years ago, most scientists and clinicians believed that the adult CNS was largely hardware, that it consisted mostly of neurons and synapses that did not change after development ended early in life. This assumption left limited scope for neurorehabilitation. When CNS injury or disease occurred, scientists and clinicians did not believe that damaged brain regions or transected spinal cord pathways could be repaired or regenerated. Thus, they sought only to maximize use of the regions and pathways that remained functional.

In fact, even 50 years ago, substantial evidence indicated that the adult CNS was plastic.^47–53 But this evidence drew little attention. The prevailing assumption that the CNS was mainly hardware was finally dispelled in the 1980’s. Merzenich et al. (1984)⁵⁴ showed that digit loss in monkeys led to reorganization of limb representation in sensorimotor cortex. And, in a seminal Science paper, Nudo et al. (1996)³⁴ provided direct evidence in a primate model that rehabilitative training after ischemic stroke caused cortical reorganization that was associated with functional recovery.

These demonstrations of plasticity in the adult CNS were followed by many more. While plasticity is most prominent during development, it continues through later life as well. Neurons change and new neurons may appear, dendrites, axons, and synapses change, glia change, hormonal influences change, even blood vessels change.^1–19 This plasticity has many mechanisms and operates on many time scales, from synaptic changes that occur in minutes, to changes in cortical representation that develop over weeks, to reflex changes that evolve over a lifetime.^34,55,56

Prior to and in parallel with recognition that CNS plasticity continues through life, came new understanding of the CNS substrates of the many skilled behaviors (i.e., skills) that are acquired and maintained through life. Skills are behaviors that enable an individual to function effectively and efficiently in the environment. They range from simple skills such as flexion withdrawal reflexes, to standard skills such as locomotion and language, to the most sophisticated athletic and cognitive performances. Skills are generally acquired through interactions with the environment (i.e., through experience). This includes even flexion withdrawal reflexes, which are acquired through experience in utero and early postnatal life.^57,58

In the middle of the 20^th century, Nikolai Bernstein made an extremely important discovery about skills. He found that a skill is not performed exactly the same way every time, even by an expert.^59,60 Kinematics vary considerably from one performance to the next. Figure 1 illustrates the kinematic variation when a skilled person uses a hammer to hit a chisel. Hammer trajectory varies from one swing to the next. Nevertheless, the hammer hits the chisel precisely every time. This precision is a key feature of the skill, an attribute that determines whether the skill is satisfactory or unsatisfactory. While kinematics may vary, the key features of a skill are maintained.⁶¹

Figure 1. A skill has key features that define it as satisfactory.

Nearly a century ago, Nikolai Bernstein recorded successive trajectories as a person repeatedly hit the head of a chisel. The figure shows multiple trajectories for each of two different chisel locations. The person was highly skilled. Nevertheless, the trajectory for a given chisel location was different every time. The only part of the trajectory that did not change was the point where the hammer hit the chisel; that was the same every time. This was the key feature of the trajectory, the position that determined whether the trajectory was satisfactory or unsatisfactory. The rest of the trajectory could vary, as long as the key feature was maintained. (Modified from^60,231.)

A flexion-withdrawal reflex has one key feature: it is satisfactory if it removes the fingertip from the hot stove very quickly; it is unsatisfactory if it does not. The particular combination of finger, hand, wrist, elbow, shoulder, torso, and leg muscle activity and joint rotations that accomplishes the withdrawal can vary. In contrast, locomotion has many key features: it is satisfactory if it is stable, symmetrical, energy efficient, etc.; it is unsatisfactory if it lacks these key features. Locomotor EMG activity and kinematics may change. But as long as key features are maintained, the skill is satisfactory. Speech also has many key features: it is satisfactory if it has appropriate word choice, correct grammar, adequate amplitude and rate, acceptable pronunciation, etc.; it is unsatisfactory if it fails to satisfy these criteria.

Bernstein’s discovery has an enormously important implication for neurorehabilitation. His discovery says that restoring a skill impaired by CNS injury or disease does not necessarily mean restoring the exact pattern of muscle activity and kinematics that produced the skill prior to its impairment. Rather it means restoring the key features of the skill, the functional attributes that make the skill satisfactory. For example, restoring the skill illustrated in Figure 1 does not mean restoring any particular trajectory, or even any specific group of trajectories. It means restoring the fact that the hammer hits the chisel precisely every time.

The studies of the past 30 years made another important discovery about skills. Skills are not like computer software, they are not stored away in special parts of the CNS and simply downloaded when needed. Instead, skill acquisition usually entails widespread CNS plasticity. For example, operant conditioning of the spinal stretch reflex (the knee-jerk reflex), or its electrical analog the H-reflex, depends on plasticity distributed from cerebellum to cerebral cortex to spinal cord (Figure 2 A). Plasticity in the brain induces and may maintain the spinal cord plasticity that directly underlies the skill.^61,62 Similarly, mastery of a finger-flexion sequence behavior depends on changes in cortex, in subcortical structures (e.g., basal ganglia and cerebellum), and in spinal cord (Figure 2 B).⁶³ Changes in the properties of neurons and/or synapses in each region contribute to the skill. Even a very simple motor skill depends on plasticity in a network of neurons and synapses that may extend from cortex to spinal cord.

Figure 2. A skill is produced by a widely distributed network of neurons and synapses.

A. The CNS substrate of operantly conditioned change in the size of the H-reflex (electrical analog of the spinal stretch reflex). The green ovals mark spinal and supraspinal sites of definite or probable CNS plasticity associated with H-reflex operant conditioning. CST, main corticospinal tract; GABA IN, GABAergic spinal interneuron; IN, spinal interneuron; MN, spinal motoneuron. Dashed pathways indicate possible intervening spinal interneurons. Open synaptic terminals are excitatory; filled ones are inhibitory; mixed may be either. The monosynaptic and probably oligosynaptic H-reflex pathway from the sensory afferents to the motoneuron is shown. Definite or likely sites of plasticity include: the motoneuron membrane (axon initial segment and axonal conduction velocity); GABAergic interneurons; GABAergic terminals and C-terminals (segmented terminals) on the motoneuron; monosynaptic afferent terminals on the motoneuron and/or their presynaptic contacts; terminals conveying oligosynaptic afferent inhibition or excitation to the motoneuron; sensorimotor cortex; and cerebellum. The data support the hypothesis that the reward contingency acts through the inferior olive to guide and maintain plasticity in the cerebellum that guides and maintains plasticity in sensorimotor cortex that (through the CST) guides and maintains the spinal cord plasticity that is largely responsible for H-reflex change. (For review:^62,74) (Updated from^14,15.)

B. CNS substrate of a motor sequence skill (i.e. a specific sequence of individual finger flexions) revealed by fMRI during skill performance. Distinct cortical, subcortical, and spinal areas are active (i.e. yellow/red areas). Activation in each area correlates positively with performance speed. Cortical activation is in contralateral sensorimotor cortex. Subcortical activation is in contralateral putamen and ipsilateral cerebellar lobules V–VI. Spinal activation is in segments C7–C8. Cortical, subcortical, and spinal activations make both overlapping and independent contributions to performance. (From⁶³.)

In recent decades, many research groups have studied the process through which the widespread network of neurons and synapses that underlies a skill is created and then changes as needed to maintain the skill. They have shown that appropriate changes (i.e., adaptations) in the network are guided by feedback during skill performance and by its outcome; these adaptations reduce errors and improve performance.^64–69 They ensure that the key features of the skill are achieved. These adaptations are an ongoing process that establishes the skill during its acquisition and maintains it through life.

The recent recognition of this ongoing process that establishes and maintains skills was presaged in the 20^th century when Bernstein’s foundational studies of skilled behaviors led him to describe them as “biodynamic structures [that] live and develop.”^59,60 The continually adapting network of neurons and synapses that underlies a skilled behavior is Bernstein’s “biodynamic structure.” It was recently given the name heksor, from the ancient Greek hexis.⁶¹ A heksor is a widely distributed network of neurons and synapses that produces a skilled behavior and changes itself as needed in order to maintain the key features of the skill, the attributes that make the skill satisfactory. These key features constitute the memory of the skill. The plasticity of the heksor ensures the stability of the memory.

The normal CNS produces numerous skilled behaviors. Their heksors overlap each other; each shares its neurons and synapses with other heksors. This overlap is most obvious in the spinal cord and its analogous brainstem areas, where motoneurons and interneurons comprise the final common pathway for essentially all skills (e.g., Figure 2 A&B). The overlap extends as well to cortex and other brain regions (⁶¹ for review). A particular neuron or synapse in cortex, cerebellum, basal ganglia, thalamus, brainstem, or spinal cord is not dedicated to one skill; it may contribute to flexion-withdrawal reflexes, locomotion, reach-and-grasp, swimming, throwing the discus, playing the violin, and many other skills. As a result, when the heksor of a new skill is created, or the heksor of an existing skill changes to maintain its key features, all the other skills that use these same neurons and synapses are affected. If these other skills are to remain satisfactory, their heksors must change as well. A heksor may change by modifying the properties of the neurons and synapses that comprise it, and it may change by incorporating neurons and synapses that were previously not part of it or by dropping neurons and synapses that were part of it.

The negotiated equilibrium concept acknowledges and begins to explore the as yet unknown process through which the many heksors that share the CNS maintain their skills throughout life.^14,61,70 According to this concept, the process is organized and executed by the heksors themselves. Their concurrent individual efforts to maintain their own skills constitute a negotiation among them; they negotiate the properties of the CNS neurons and synapses that they all use. Through this process, they establish and maintain an equilibrium satisfactory to all of them. While this process may change the CNS activity and muscle activations that underlie a skill, it maintains the skill’s key features, the features that make the skill satisfactory. Thus, the key features of locomotion include attributes such as right/left step symmetry, upright posture, adequate balance, and acceptable metabolic cost. They do not include a defined sequence of muscle contractions. Muscle activations can change, as long as the key features of locomotion are maintained. This negotiation among heksors is ongoing, it continues to maintain a negotiated equilibrium as new skills are acquired, and as growth and aging, trauma and disease, and other life events change the CNS itself and the peripheral sensory and motor structures through which the heksors interact with the body and the world.

Figure 3 A illustrates heksor negotiations and their results. Figure 3 B–D provides a simple example of a negotiation between a new heksor and an existing (i.e., old) heksor in a rat. The new heksor created by operant conditioning of the right soleus H-reflex includes the pathway of the soleus spinal stretch reflex (Figure 2 A), which is already part of the old locomotion heksor. By weakening (down-conditioning) or strengthening (up-conditioning) this pathway on the right side, the new heksor changes right ankle angle during locomotion. If this were all that occurred, it would impair a key feature of locomotion – right/left symmetry in hip height. The right/left asymmetry in hip height would twist the spine, and the rat would tilt to right or left as it walked. But this does not happen because the locomotion heksor changes itself in response to the new H-reflex heksor. It changes so that an opposite change in right hip angle compensates for the change in ankle angle; the result is that locomotor EMG and kinematics change so as to maintain the key feature of hip height symmetry. The new and old heksor have negotiated an equilibrium satisfactory to both of them.

Figure 3. Heksors create and maintain a negotiated equilbrium of CNS neuronal and synaptic properties.

A. Heksor negotiations through life. The first image is a naïve CNS. The triangles represent the many CNS neurons or synapses; their properties may undergo activity-dependent plasticity. The second image is the prenatal and early postnatal CNS, which acquires a flexion-withdrawal reflex skill. Its heksor comprises neurons and synapses (green) with properties that have changed to produce the new skill. The third image is a little later in life, when the skill of locomotion is acquired. Its heksor (red) overlaps the flexion-withdrawal heksor. The heksors negotiate the properties of the neurons and synapses that they share so that each maintains the key features of its skill. This negotiation may also affect their unshared neurons and synapses as well. Thus, the properties of each neuron or synapse in either heksor may be affected by both of them. The fourth image is later in life when the athletic skill of throwing the discus is acquired. Its heksor (orange) overlaps the other two. The three heksors negotiate the properties of their shared neurons and synapses; this may also affect those they do not share. In sum, the properties of neurons and synapses in each of the three heksors may be affected by all three of them. Together, the heksors keep the CNS in a state of negotiated equilibrium that maintains the key features of the skill produced by each of them. (From⁶¹.)

B. A rat with chronically implanted soleus EMG electrodes walks on a treadmill. The traces show the right and left soleus EMG bursts that support the right and left stance phases of the step cycle. (Modified from²³².)

C and D. Soleus H-reflex conditioning changes how locomotion is produced.

C. H-reflex conditioning affects locomotor EMG. Average absolute value of right soleus locomotor EMG bursts before (continuous black) and after (dotted blue or red) right soleus H-reflex conditioning in a down-conditioned rat (left, blue) and an up-conditioned rat (right, red). The H-reflex pathway contributes to the locomotor burst. Thus, the burst is decreased by down-conditioning and increased by up-conditioning. (Modified from²³².)

D. These EMG changes affect locomotor joint angles, but they do not impair locomotion. Center: Right stance-phase anterior-ankle, posterior-knee, and anterior-hip joint angles as a rat walks on the treadmill. Left (left to right): The first image shows the angles before down-conditioning of right soleus H-reflex. The second shows that the weaker soleus burst (e.g. E, left-side) reduces ankle angle. This alone would reduce right hip height and cause right/left asymmetry in hip height (i.e., the rat would tilt to the right and the spine would be twisted). This does not happen. The third image shows that a concurrent increase in hip angle compensates for the decrease in ankle angle, so that right hip height does not change. The joint angles change, but the key feature, right/left symmetry in hip height, is maintained. Right (left to right): The first image shows the angles before up-conditioning of right soleus H-reflex. The second shows that the increased soleus burst (C, right-side) increases ankle angle. This alone would increase right hip height and cause right/left asymmetry in hip height (i.e., the rat would tilt to the left and the spine would be twisted). This does not happen. The third image shows that a concurrent decrease in hip angle compensates for the increase in ankle angle, so that right hip height does not change. The joint angles change, but the key feature, right/left symmetry in hip height, is maintained. (Modified from²³³.)

The example in Figure 3 is a tiny piece of an enormously complex process that continues through life. Each of the many heksors that share the CNS is continually changing itself to maintain its own skill despite the creation of new heksors, despite concurrent changes in existing heksors, and despite growth, aging, and other life events. The concurrent changes of all the heksors keep the CNS in a state of negotiated equilibrium in which all their skills are consistently satisfactory. This equilibrium is similar to a Nash equilibrium as defined in game theory.^71–73 The players – all the heksors – negotiate an equilibrium in which none can better its performance by changing further. In the healthy CNS, this negotiated equilibrium ensures that each skill is satisfactory. The importance of heksor negotiation in maintaining satisfactory CNS function is illustrated by recent evidence that it affects the properties of the axon initial segment (AIS) of the spinal motoneuron, the structure that ultimately determines whether and when muscles are activated.⁷⁴

These two concepts – heksor and negotiated equilibrium – comprise a new paradigm that is able to explain how skilled behaviors are acquired and maintained in a ubiquitously plastic CNS. Exploration of the mechanisms that enable heksors to retain the key features of their skills, and to interact with other heksors to produce a negotiated equilibrium is just beginning.^61,74 Nevertheless, even at this early time, these concepts and the appreciation of CNS plasticity on which they are based redefine the primary goal of neurorehabilitation and generate new therapeutic strategies that are already providing remarkable pre-clinical and initial clinical results.

Technological Advances

The scientific advances summarized above occurred in parallel with technological advances that increase the potential therapeutic impact of the new science. Stated most simply, the new science shows that the adult CNS is plastic; the new technologies provide clinical access to this plasticity. They make it possible to use this plasticity to restore skills impaired by injury or disease. These new technologies fall into three classes: biological and bioengineering; neurostimulation; and computer-based real-time interactive training. Many therapies that use these technologies are at various stages on the multi-step pathway from basic research to clinical dissemination.

New biological agents can enhance neuronal and/or glial plasticity. Intraspinal transplants of Schwann cells, mesenchymal stromal cells, olfactory ensheathing cells, or oligodendrocyte precursor cells can protect neurons, reduce glial scarring, increase vasculature, and produce axonal growth, re-myelination, and functioning synapses.^75–85 Several of these agents have undergone Phase-1 clinical trials.^86–88 Growth factors (e.g., nerve growth factor, neurotrophin-3, brain-derived neurotrophic factor, glial cell-derived neurotrophic factor) administered to the damaged spinal cord by injection, release from biomaterials, transplanted genetically modified cells, or genetic modification of spinal cord cells can protect neurons and produce axonal regeneration.^89–91 Agents that interfere with the endogenous axon growth-inhibitory molecules present in the damaged spinal cord (i.e., chondroitin-sulfate proteoglycans (CSPGs), reticulon 4 (Nogo)) enable axon regrowth.^92–94 With most of these agents (but not all⁸⁵), axons do not grow long distances or establish functional synaptic connections with host spinal cord neurons. Much current research focuses on overcoming these limitations.

Bioengineering is a rapidly growing component of neurorehabilitation research.^95,96 Solid or injectable biocompatible materials can provide platforms that enhance the survival of transplanted cells.^97–99 These materials can provide targeted delivery of agents that stimulate neural repair and functional recovery.^91,100–102 While biomaterials are often combined with cell transplants or growth factors, some can on their own trigger tissue formation, angiogenesis, and axonal regeneration.^103–105

Several kinds of brain, spinal, or peripheral stimulation can increase plasticity by augmenting existing capacities for plasticity or providing additional capacities (e.g., vagal nerve stimulation (VNS)^106,107). After SCI eliminates effective supraspinal control of spinal motoneurons, invasive or noninvasive spinal cord stimulation can enable these motoneurons to respond to the weak corticospinal influence that remains and thereby participate in skill-specific practice that leads to wider beneficial plasticity.¹⁰⁸ Transcranial direct current stimulation (tDCS) via scalp electrodes can modulate cortical excitability and may augment the efficacy of skill-specific practice.^109,110 Appropriate pairing of peripheral nerve stimulation with movement-related EEG potentials can enhance recovery after stroke (Reference ¹¹⁴)

The ubiquitous plasticity that occurs in the CNS through life is guided by interactions with the environment (i.e., by experience). Rehabilitation robots can standardize and shape these interactions to promote beneficial plasticity.¹¹¹ The rapid development and widespread availability of computer hardware and realtime software has enabled creation of training protocols that target beneficial plasticity. Most of these protocols are noninvasive. As described in later sections, they enhance skill-specific practice, and lead to wider beneficial plasticity.^{112,113,155,157}

Neurorehabilitation Today

The newly recognized plasticity of the CNS and the appearance of techniques and technologies that can encourage and shape that plasticity introduce the theoretical possibility that skills lost to CNS injury or disease, such as locomotion or speech, might be largely, or even fully restored. The new science clarifies the goal of neurorehabilitation and reveals the full complexity of the problem that must be solved to achieve that goal.

The Goal

When The CNS is damaged, neurorehabilitation need not – and in fact cannot – restore the CNS to its pre-injury state. Because its ongoing interactions with the internal and external environments change the CNS continually through life, it seldom – if ever – returns to a previous state. Complex patterns of plasticity in CNS neurons and synapses, whether physiological or pathological, cannot simply be undone or erased. Nevertheless, this does not prevent restoration of skilled behaviors.¹¹⁵ To maintain their skills, heksors adapt early in life as growth proceeds; they adapt through life as the new heksors underlying new skills are created; and they adapt later on to mitigate the effects of aging. Thus, the neural activity that produces locomotion in a healthy 50-year-old is probably not the same as the neural activity that produced locomotion when that person was a healthy 20-year-old. Similarly, when CNS trauma or disease occurs, damaged heksors may adapt to restore their skills. The CNS activity underlying a restored skill may be different, but the skill can still be intact if its key features, the attributes that make it satisfactory, are still achieved. Only the details (e.g., the individual muscle contributions and the kinematics) may be different.

We note that this new definition of recovery differs to some extent from that presented by Levin et al. (2009¹¹⁶); their definition stresses recovery of premorbid movement patterns, which may differ across individuals. The new definition focuses simply on recovery of the key features of a skill, which define it as satisfactory. When its key features are restored, the skill can be considered fully functional in clinical terms. Operationally, this definition of recovery might be reduced to the following: An expert (e.g., physical therapist, physiatrist), with no knowledge of a person’s premorbid level of function or history of neurological impairment, observes all aspects of the person’s performance of a previously impaired skill (e.g., locomotion or reach and grasp with a specific arm) and concludes that the skill is satisfactory. (It should be acknowledged that this definition of recovery – while clinically useful – does not take into account a person such as a formerly elite runner who after a stroke and subsequent treatment, runs satisfactorily, but no faster than an average person. The patient may well be dissatisfied with this recovery.)

From this new definition of functional recovery follows the primary goal of neurorehabilitation. When a skill such as locomotion, reach-and-grasp, or speech is impaired by CNS damage, the goal is to enable the damaged heksor responsible for the skill to repair itself so that it can once again maintain the key features of the skill, and thereby improve the functional status of the individual. The neural activity that produces the restored skill may – and probably will – differ from the activity that produced the skill prior to impairment; just as the neural activity underlying locomotion differs with age. Nevertheless, the skill can still be satisfactory, and the heksor can still establish with other heksors a negotiated equilibrium in which all the heksors maintain their skills. In sum, CNS function can be restored. Over 20 years ago, a task force for the Agency for Healthcare Research and Quality, Center for Outcomes and Effectiveness Research proposed that functional status be considered the “sixth vital sign.”¹¹⁷ Today, as NIH proceeds to update its mission statement¹¹⁸, we believe that the entire neurorehabilitation community would strongly support the underlined addition to the statement: “To seek fundamental knowledge about the nature and behavior of living systems and to apply that knowledge to optimize health, optimize function, and prevent or reduce illness for all people.”

Because the goal is to restore the key features of impaired skills, standard laboratory or clinical measures, such as reflex amplitudes, are important or meaningful only in how they affect or reflect the key features of important skills.¹⁸⁸ The H-reflex or motor evoked potential (MEP) operant conditioning that improves walking after incomplete SCI or arm/hand function after stroke may well make these physiological measures larger or smaller, and may create right/left asymmetries in them. Nevertheless, if the conditioning restores an important skill such as walking (e.g., if it restores right/left symmetry¹¹²), these associated changes need not be a concern.

This principle might be best understood in the context of the wide variations in physiological measures found in people who are entirely intact neurologically. Knee-jerk reflexes may be very large in one person and difficult to detect in another. Such variations reflect genomic and musculoskeletal differences between the individuals, as well as differences in the roster of skilled behaviors each has acquired through life (e.g.,^189,190) and in the equilibrium of CNS properties that the heksors underlying each person’s skills have negotiated. Similarly, in a person who has regained skills impaired by CNS injury, physiological measures are likely to reflect the plasticity through which damaged heksors regained the key features of their skills. The uncertain significance of individual physiological measures is indicated also by the overlap seen in such measures between people without CNS injury and people with significant disabilities.^188,191 In sum, measures such as spinal reflexes, cortical evoked potentials, and fMRI topographies have meaning only in the context of the current state – satisfactory or not satisfactory – of the individual’s skills.

The Problem

The time after a stroke or other CNS trauma is usually divided into acute, subacute, and chronic stages.⁴⁰ The acute and subacute stages, which typically last about 6 months, are a time of rapid changes, both harmful and beneficial, as the effects of the injury continue to develop and spontaneous repair processes get underway. The initial days and weeks are often further complicated by extra-CNS problems and therapies (e.g., cardiovascular insufficiencies and orthopedic procedures). In contrast, the chronic (>6 months post-injury) period is an ostensibly stable time in which individuals are left to deal with the disabilities created by the injury, typically for the rest of their lives. As stated in the Introduction, this article focuses on the unprecedented opportunities now provided by the new science and the new technologies. In accord with this focus, the article now addresses the potential impact of these new opportunities on function in the chronic phase of injury, the phase in which 20^th-century rehabilitation had very little to offer.

The CNS in the chronic state after a stroke, an SCI, or other structural lesion, is often conceived as the healthy CNS was in 1970, that is, as a largely hardwired structure with little capacity to change. That concept is now as obsolete for the damaged CNS as it is for the healthy CNS. The plasticity of the damaged CNS is likely to be even more complex than that of the healthy CNS. In the healthy CNS, plasticity is driven mainly by interactions with the body and the world. Heksors change as needed to maintain a smoothly functioning negotiated equilibrium that ensures the performance of all skilled behaviors. In the chronically damaged CNS, plasticity is also driven by interactions with the body and the world, but it also includes ineffective or even counterproductive changes produced by the efforts of damaged heksors to restore their skills. Furthermore, the damaged CNS is subject to additional plasticity due to factors not present in the healthy CNS.

One factor is diaschisis. As originally defined by Monakow in 1914, diaschisis meant depression of function in CNS areas connected to a lesioned area.¹⁶² Recent histological and imaging studies have extended the term to include substantial neuronal loss, astroglial effects, and microvascular damage.^163–166 The advent of functional imaging has further developed the term to include the complex changes that a lesion in one area can produce in the activity of the multiple functional networks that connect cortical and subcortical areas.^167–178

Another factor contributing to plasticity in the damaged CNS is the reemergence of interactions that had been resolved during CNS development early in life. For example, the reticulospinal pathways that develop relatively early enable creation of heksors that produce the generalized flexion/extension movements prominent in infancy. As development continues and corticospinal tract pathways begin to function, these primitive flexion/extension heksors are subsumed within the highly differentiated heksors enabled by corticospinal pathways.^179,180 A stroke or other injury that damages sensorimotor cortex can release these primitive heksors, which may partially restore skills lost to the injury.^181,182 However, if and when cortical function begins to recover, these primitive heksors may conflict with the highly differentiated heksors that incorporate corticospinal pathways.^182,183

The re-emergence of developmental-like processes may also contribute to the complex interactions that occur between a cerebral hemisphere damaged by a stroke or other unilateral injury and the contralateral, overtly intact hemisphere. Early in life, activity-driven competition between contralateral and ipsilateral CST pathways establishes the normal dominance of the contralateral CST; this dominance is henceforth maintained by ongoing activity through life.¹⁷⁹ However, if a unilateral stroke reduces contralateral CST activity, the ipsilateral CST may become abnormally strong.¹⁸⁴ This might be beneficial or harmful. What happens will be determined in large part by callosal-based interactions between the hemispheres.^185–187

In addition to plasticity due to damaged heksors, diaschisis, reemerged primitive heksors, and multilevel ipsi-contra interactions, the damaged CNS may undergo plasticity due to creation of new heksors that compensate for those that have been impaired (e.g., new heksors that control a wheelchair, eye-movement based communication, a brain-computer interface (BCI), or previously unused muscles¹⁶⁰).

Finally, in people with chronic CNS impairments, the extent and characteristics of disability may vary substantially across days, weeks, and months due to changes in the nature and level of daily activities, in living conditions and social interactions, in other health issues, and in other as yet undefined factors.

In summary, the chronically damaged CNS is likely to be a contentious, even chaotic environment in which many agents – damaged heksors, primitive heksors, compensatory heksors, diaschisic processes, ipsi-contra influences, life events – interact continually, with results often far less than optimal. Thus, any rehabilitation therapy is venturing into a complex, active, and perhaps even hostile environment.

New Therapeutic Strategies

Since the turn of the century, neurorehabilitation has focused on skill-specific practice – practice of locomotion, reach-and-grasp, language, and other impaired skills. This strategy can engage the substantial plasticity and learning mechanisms present in the adult CNS. But it often fails to restore full function, for two reasons.

The first and most obvious reason why skill-specific practice alone is often not fully effective is that some impairments cannot be remedied by the capacities for plasticity available in the adult CNS. Transected spinal cord pathways or lost areas of cerebral cortex do not spontaneously regenerate and reestablish appropriate and effective connections with other CNS areas. These phenomena require mechanisms of plasticity akin to those responsible for CNS development early in life when CNS cell types and areas arise and their connections are established (e.g.,¹¹⁹).

The first new strategy addresses this problem by increasing plasticity in general. New biological agents and bioengineering technologies can enhance the mechanisms of plasticity already available or provide additional mechanisms that are not normally present in the adult CNS. By providing a damaged heksor with more options for plasticity, these interventions enable it to change in ways that were not previously possible. For example, a therapy that promoted regrowth of CNS axons severed by an SCI and enabled them to reestablish effective synapses (e.g.,⁸⁵) could enhance the efficacy of skill-specific practice, thereby leading to beneficial plasticity throughout the damaged heksor. The animal and human studies reviewed below provide striking evidence of these wide effects.

The second reason why skill-specific practice alone is not sufficient is less obvious but also important. Skill-specific practice is not sufficiently specific. The heksors underlying skills such as locomotion are extremely complex; plasticity can occur at many sites and by many mechanisms. When a heksor is damaged, nothing ensures that the available mechanisms of plasticity are optimally engaged in the optimal sequence at the optimal sites in the damaged heksor. Overall direction does not exist; potentially beneficial mechanisms may not be engaged at all (e.g., pg. 20 in⁶¹). Thus, skill-specific practice often fails to produce the best possible pattern of plasticity, and recovery is not maximized.

The second new strategy addresses this problem of insufficient specificity by targeting beneficial plasticity to a critical site in a damaged heksor. It thereby engages an available repair option that the damaged heksor had not engaged. The targeted plasticity enables better skill-specific practice, which produces beneficial plasticity elsewhere in the damaged heksor. For example, operant up-conditioning of the tibialis anterior motor evoked potential (MEP) in a person with SCI can reduce footdrop; it thereby enables better locomotor practice that leads to wider beneficial plasticity.¹⁴³ Furthermore, the targeted beneficial plasticity may improve the subsequent adaptations of the damaged heksor; after the protocol ends, the heksor may continue to use the repair option that the protocol engaged.^61,121 In accord with Bernstein’s description, the heksor is a “biodynamic structure that lives and develops.”⁶⁰ In short, appropriate change at a critical site leads to beneficial plasticity throughout the heksor. The studies reviewed below demonstrate these wider effects and their lasting functional impact.

Preclinical and clinical research groups are currently testing a variety of therapeutic protocols based on one or the other of the new strategies. These efforts are proliferating as the therapeutic efficacy of the strategies becomes more and more apparent. A new protocol often begins in animals and, if results are favorable, moves into humans. Alternatively, a minimal-risk protocol may begin in humans and, if results are positive, lead to animal studies to elucidate mechanisms or optimize outcomes. Animal and human studies may continue in parallel and serve each other; animal studies justify and guide human studies, and human studies provide insights that guide further animal studies. As a result, neurorehabilitation research is becoming a synergistic and iterative combination of laboratory and clinical endeavors.¹²² The next two subsections describe representative examples of promising therapies that implement the “generalized-plasticity” or “targeted-plasticity” strategy and are now in various stages of development.

The Generalized Plasticity Strategy

Repairing Nervous Tissue

SCI destroys nervous tissue and impairs motor, sensory, and autonomic function. The primary injury is typically followed by secondary tissue loss that often creates large cystic cavities at the site of the primary injury. Endogenous nervous tissue repair is very limited; thus, the prognosis for functional recovery is poor. Cell transplantation is under study as a means of eliciting nervous tissue repair and/or replacing lost neural cells.^75–84,123 Schwann cells myelinate peripheral nerves, and are therefore critically important for peripheral nerve repair. They secrete many different molecules that may also support CNS repair. In animal models of SCI, Schwann cell transplantation increases neuronal survival, encourages axonal growth and myelination, promotes blood vessel formation, reduces glial scarring, benefits immunomodulation, and improves recovery of function.^124,125 These preclinical studies have prompted clinical trials that transplant autologous Schwann cells into people with SCI.^86,87 Figure 4 A illustrates Schwann cell transplantation and some of its effects on tissue repair and functional recovery after SCI.

Figure 4. Generalized Plasticity Therapies I.

A. Schwann cell (SC) transplantation in spinal cord injury. Transplantation of SCs into an adult rat spinal cord contusion improves sensorimotor function.

1. Schematic of a typical experimental SC transplantation procedure. The rat sciatic nerve is removed and processed to isolate SCs for allogeneic transplantation. The cells are cultured through several passages to obtain proper amounts for injection into the contused spinal cord. Typically, following SC transplantation, rats are tested weekly for sensorimotor motor function using standardized tests until the spinal cord is processed for histology and imaging. Autologous SCs have been transplanted safely into people with sub-acute and chronic spinal cord injury.

2. SC transplant in an adult rat spinal cord contusion injury. Prior to transplantation, SCs are often genetically modified to express green fluorescent protein, which enables detection after injection and studies of transplant survival and functioning.^234,235

3. A SC transplant limits nervous tissue loss in the contused spinal cord segment.²³⁴ Rats that received a SC transplant in their contused spinal cord (green bars) have significantly more spared nervous tissue than rats that had a control injection with culture medium (orange bars). SCs secrete molecules (i.e., growth factors and cytokines) that increase plasticity in the nervous tissue surrounding the injury epicenter. This is a key aspect of SC transplant-mediated spinal cord repair because it can lead to tissue protection and tissue (re)generation that can contribute to functional improvements.

4. Hindlimb overground walking is improved in rats transplanted with SCs in a contusion site in the thoracic (T9) spinal cord (e.g.,^234,236). Overground walking ability assessed using the Basso-Beattie-Bresnahan (BBB) scale²³⁷ was significantly improved in rats that received a SC transplant (gold bars) compared with rats that had a control injection with culture medium (blue bars). The walking improvements reached significance at 6 and 8 weeks after cell transplantation. While a SC transplant can promote axon growth and myelination,²³⁸ it is presently not clear how this axon response is involved in the observed hindlimb functional improvements. It is possible that SC-mediated events affect neuroplasticity in the spinal cord levels below the injury site and so contribute to functional recovery.

5. A SC transplant in a contusion site in the cervical (C5) spinal cord improved forelimb reach and grasp function.¹²⁴ A significant improvement in this function was found in rats with a SC transplant (green bars) compared with control rats injected with culture medium (orange bars). Reach and grasp function was tested using a Montoya staircase in which rats reached and grasped sugar pellets. The improved reach and grasp in rats that received the SC transplant were significant at 4 and 8 weeks after cell transplantation.

B. Implantation of nanofiber hydrogel composite (NHC) in the contused spinal cord. Injection of the composite into an adult rat spinal cord contusion injury facilitated the formation of tissue in the contusion site.¹⁰⁴ The components of NHC can be mixed just prior to injection into the spinal cord. Alternatively, NHC can be made, fragmented, and stored to be used for injection at later times. Injected rats are usually tested for motor function and, at the end of the experiment, their spinal cords are processed for histology and imaging (see schematic in panel A1).

1. An NHC transplant promotes the formation of new blood vessels (angiogenesis). Injection of NHC into a contusion injury site in the adult rat spinal cord (outlined, dashed line) resulted in the formation of tissue with newly formed blood vessels.^99,104 The blood vessels can be recognized using immunostaining with antibodies against RECA-1 (red). The blood vessels are clearly visible in an enlargement of the outlined rectangle in the center of the contusion in the top right corner of panel B1. Vascularization of the tissue is important because the blood vessels provide oxygen and nutrients that maintain its health.

2. Bar graph showing that significantly more blood vessels are present in a contusion site injected with NHC compared with a contusion site injected with PBS (control). One of the components of NHC is hyaluronic acid; this may be the component that facilitates the angiogenesis process (adapted from¹⁰⁴).

3. Axons can be found in NHC-facilitated tissue in a contusion. Many more axons positive for neurofilaments (green) are present in a contusion site transplanted with NHC (bottom) than in a contusion site injected with PBS (top).^99,104 Axons are often associated with the blood vessels that are found in the NHC-injected contusion site.

4. Bar graph showing the significantly higher axon density in a contusion site injected with NHC (green bar) compared with a contusion site injected with PBS (control; blue bar) (adapted from¹⁰⁴).

5. A NHC-transplanted contusion site contains neuron-like cells. Immunostaining with antibodies against Tuj1 was used to recognize neuron-like cells.¹⁰⁴ The presence of these neuron-like cells, and the neurofilament positive axons, reflect the NHC-mediated neuroplasticity events within the contusion site. Tuj1 immunostaining is greater in the NHC-transplanted rat (right) than in the Control rat (left).

6. More neuron-like cells are present in an NHC-treated contusion site. Tuj1 immunostaining is significantly greater in contusion sites transplanted with NHC (gold bar) than in contusion sites injected with PBS (control; blue bar).

7. Transplantation of NHC did not affect hindlimb walking ability as assessed with the BBB scale.¹⁰⁴ There were no significant differences in walking performance between rats injected with NHC (green bars) compared with control rats injected with culture medium (orange bars). While NHC facilitates the formation of tissue in a spinal cord contusion site, it was not shown to increase motor function. Future efforts may focus on combining the neuroplasticity enhancing properties of NHC with other complementary neuroplasticity-enhancing approaches such as Schwann cell transplantation (Figure 4 A) or daily ISP treatments (Figure 5 A).

Providing a platform for regeneration

The site of an SCI is typically cytotoxic and therefore inhospitable to transplants of repair cells.¹²⁶ Injection of nanofiber hydrogel composite (NHC) may reduce these obstacles to plasticity and regeneration.^103,104 NHC injection into the contused spinal cord in the subacute post-injury period leads to formation of tissue in the contusion site that has a pro-regenerative inflammatory profile and a high density of blood vessels and axons; it thereby creates a pro-regenerative inflammatory milieu.¹⁰⁴ Figure 4 B illustrates and quantifies these notable effects. While NHC alone may not improve functional recovery (Figure 4 B7), it could provide a hospitable platform for cell transplants that do improve function (e.g., Figure 4 A).

Enabling regrowth of severed axons

After an SCI, CSPGs produced mainly by reactive astrocytes bind to receptors on the axon growth cone and inhibit growth.^127–132 An Intracellular Sigma Peptide (ISP), also known as NVG-291, prevents this inhibition.^127,129 In animals with SCI, subcutaneous or intraperitoneal ISP treatment can promote regeneration of corticospinal and serotonergic axons important for locomotion, enhance remyelination, release CSPG digesting enzymes, activate growth-associated pathways, and improve the immune response; these effects are accompanied by better recovery of locomotor, respiratory, and urinary function.^133–141 These preclinical studies indicate that ISP treatment may be a new minimally invasive method for increasing beneficial plasticity and enhancing functional recovery in people with CNS injury. Phase-1 clinical studies in people with chronic SCI are underway. Figure 5 A illustrates the use, mechanism of action, histological effect, and remarkable functional impact of ISP therapy in some rats, though not in all.

Figure 5. Generalized Plasticity Therapies II.

A. Promoting neuroplasticity in the injured spinal cord can improve sensorimotor function after spinal cord injury. Injury-mediated axonal damage is a major contributor to the loss of function following spinal cord injury. Endogenous axon regeneration in the injured spinal cord is poor. Enhancing plasticity in the nervous tissue after spinal cord injury may promote axonal regeneration that improves function.

1. Schematic of an adult rat receiving daily systemic intraperitoneal injections with intracellular sigma peptide (ISP). ISP is also known as NVG-291, which is currently produced by NervGen Pharma. Besides promoting axon growth, ISP may also stimulate axon remyelination, which can further contribute to the overall repair process.

2. Schematic of ISP interfering with the intracellular components of the signaling pathway of the protein tyrosine phosphatase sigma (PTPσ) receptor following the receptor’s binding to chondroitin sulfate proteoglycan (CSPG). CSPGs are well-known axon growth-inhibiting molecules that are abundantly expressed in a contused spinal cord segment.

3. A spinal cord contusion injury causes a decrease in serotonergic axon innervation in the spinal cord caudal to the impact site. Serotonergic axons are known to be involved in the execution of motor actions; promoting their growth may lead to functional improvements. With permission, from²³⁹.

4. Daily ISP treatment to adult rats with a spinal cord contusion injury resulted in abundant serotonergic axon sprouting below the level of the contusion injury. This response reflects the enhanced neuroplasticity due to ISP treatment. With permission, from²³⁹.

5. Daily ISP treatment results in improved hindlimb overground walking as assessed with the BBB scale. (ILP is less effective.) In some rats, the ISP treatment resulted in functional improvements (responders), but not in all rats (non-responders). The mechanistic underpinnings of this difference between rats in their response to ISP treatment are unknown. With permission, from²³⁹.

6. Sensorimotor function was improved in rats that received daily ISP injections (ISP) compared to rats that got daily vehicle injections (vehicle controls). Sensorimotor function was tested on a horizontal ladder; foot slips from the rungs were quantified. Note that fewer foot slips reflects better sensorimotor performance. It was also clear with this test that ISP treatment improved in some rats (Responders) but not all rats (Non-Responders). With permission, from²³⁹.

B. Vagus nerve stimulation (VNS) coupled with skill performance can improve recovery of sensorimotor function after CNS injury.

1. VNS (left image) given after good skill performance enhances skill recovery through cholinergic reinforcement that results in skill-appropriate modulation of neurons in primary motor cortex. With permission from ¹⁰⁷

2. Top: Changes in Fugl-Meyer Assessment-Upper Extremity (FMA-UE) score from Baseline to Day 1 (left) and from Baseline to Day 90 (right) after end of therapy. Bottom: Changes in Wolf Motor Function Test-Functional (WMFT) score from Baseline to Day 1 (left) and from Baseline to Day 90 (right) after end of therapy. Circles are mean group values and vertical lines indicate 95% confidence intervals for the Control (blue) and VNS (red) groups. * indicates p<0.05 for the between group difference. Note that the VNS group appears to continue to improve after therapy ends. With permission from ¹⁰⁶

3. VNS enhances recovery of function in pre-clinical animal studies and in clinical trials. The control-group improvement found in each study is shown as 1.0. Across these human and animal studies, VNS enhances recovery 1.8 to 2.9 times. With permission from ¹⁰⁶

Increasing the plasticity produced by skill-specific practice

Over the past decade, animal studies and clinical trials have shown that vagal nerve stimulation (VNS) in conjunction with skill-specific practice enhances functional recovery after stroke.¹⁰⁶ As assessed by a variety of measures, the improvement with VNS is typically 1.8–2.9 times that achieved without VNS. Recent mechanistic studies in mice indicate that vagal afferents act through the nucleus tractus solitarius to increase cholinergic neural activity in basal forebrain that in turn modulates neuronal activity in primary motor cortex.¹⁰⁷ Figure 5 B summarizes this putative mechanism, its impact, and the functional impact of VNS in people with chronic stroke. While VNS is a generalized-plasticity therapy, its efficacy depends on its temporal pairing with active movement.^107,264,265 This pairing targets the plasticity enabled by VNS to the CNS circuitry responsible for that movement.

The Targeted Plasticity Strategy

The value of noninvasive targeted plasticity was first shown in rats in which a unilateral incomplete spinal cord injury had weakened stance on that side, causing right/left asymmetry in the step cycle (Chen et al. 2006b, 2014).^120,121 A variety of targeted-plasticity protocols are under study in people. Here we review protocols that target spinal plasticity, corticospinal plasticity, or cortical plasticity.

Operant conditioning of spinal reflexes and other evoked responses

Operant conditioning of the spinal stretch reflex (SSR), initially demonstrated in monkeys in 1983, was first demonstrated in humans by Wolf and Segal’s lab in 1989.^262,263 Therapeutic applications of this phenomenon have focused on operant conditioning of the H-reflex, the electrical analog of the SSR.⁶² In a controlled study of people with impaired locomotion caused by chronic incomplete SCI, Thompson et al. (2013) found that down-conditioning the soleus H-reflex in the more spastic leg produced faster and more symmetrical locomotion.¹¹² Participants said that they were walking faster and farther in their daily lives; and some reported reduced clonus, easier stepping, and other gains. Manella et al. (2013) also reported positive results with this new therapy.¹⁴² By rewarding a smaller H-reflex, the down-conditioning protocol creates what is essentially a therapeutic heksor; this heksor has only one key feature, a smaller H-reflex. Because the locomotion heksor also uses the down-conditioned reflex pathway, it is a subsidiary beneficiary.

Figure 6 A illustrates two crucial aspects of this noninvasive therapy. First, the targeted plasticity triggers much wider beneficial plasticity. The faster walking and restored right/left symmetry reflect bilateral plasticity; locomotor EMG activity is better in multiple muscles of both legs. We hypothesize that this occurs because the targeted plasticity improves walking practice (e.g., by reducing clonus)^143–145; and better practice improves the concurrent adaptations of the damaged locomotor heksor, thereby producing wider beneficial plasticity. Second, the benefits do not wane after conditioning ends; they persist (Figure 6 A2). It appears that the ongoing adaptations of the restored heksor maintain these beneficial changes, just as the heksor’s adaptations had maintained satisfactory locomotion before the SCI. Animal data suggest that, after conditioning ends, the locomotion heksor continues to change the H-reflex; it thereby accrues further benefits.^61,121

Figure 6. Targeted Plasticity Therapies.

A. Operant conditioning of a spinal reflex. Down-conditioning of the soleus H-reflex in the more impaired leg improves locomotion bilaterally in people with chronic incomplete SCI.

1. Average soleus H-reflex in a person with SCI before (solid) and after (dashed) successful soleus H-reflex down-conditioning. The H-reflex is smaller after down-conditioning. The M-wave (i.e., the direct muscle response), which indicates the effective stimulus strength, does not change. A small stimulus artifact is evident.

2. Soleus EMG activity during locomotion measured before and after soleus H-reflex down-conditioning and then 6 months after conditioning ended in a person with chronic incomplete SCI. Down-conditioning greatly reduced soleus EMG activity during the swing phase of locomotion. Thus, it restored a more normal pattern of soleus locomotor activity. This beneficial change remained 6 months after conditioning ended.

3. Bilateral soleus and tibialis anterior (TA) EMG activity during walking before (dashed) and after (solid) down-conditioning sessions in another person with SCI. The step cycle is divided into 12 equal bins, starting from foot contact. Thus, bins 1–7 are the stance phase and bins 8–12 are the swing phase. After successful down-conditioning reduced H-reflex size in the more impaired leg, the soleus stance burst was increased and the concurrent inappropriate stance-phase TA activity was decreased bilaterally. Down-conditioning the soleus H-reflex in the more impaired leg improved locomotion in both legs.

4. (Top) Ten-meter walking speed (mean(±SE) in % of pre-conditioning speed) and step-cycle symmetry (mean±SE) before and after successful down-conditioning in six people with chronic SCI. Step-cycle symmetry was measured as the ratio of the time between the nonconditioned leg’s foot contact (nFC) and the conditioned leg’s foot contact (cFC) to the time between cFC and nFC. Thus, a perfectly symmetrical gait has a value of 1. Down-conditioning increased speed by an average of 59% and largely restored gait symmetry (p<0.05 by paired t-test for both changes). (Bottom) Successive step cycles in a person with chronic SCI before and after successful down-conditioning of the soleus H-reflex in the more impaired leg. Each nFC (solid circle) and cFC (open circle) are shown. The short vertical dashed lines mark the midpoints between nFCs (i.e., the midpoints of the step cycle), which is when cFC should occur. Before H-reflex down-conditioning, cFC occurs too late; after successful down-conditioning, it occurs on time. Gait symmetry is restored. (A1–4 modified from¹¹².)

B. Paired associative stimulation (PAS) improves motor function in people with chronic incomplete SCI.

1. Average MEPs from biceps brachii and abductor pollicis brevis muscles of representative participants before (pre) and after (post) 10 sessions of exercise plus PAS or Sham PAS. MEPs in both muscles increased with PAS, but not with Sham PAS.

2. Maximum voluntary contractions (MVCs) (rectified EMG activity) from biceps brachii and first dorsal interosseous muscles of representative participants before (pre) and after (post) 10 sessions of exercise plus either PAS or Sham PAS. MVCs increased with PAS, but not with Sham PAS.

3. (Top) Average (±SD) MEPs (left) and MVCs (right) of all muscles tested for participants before (PRE) and after 20 (POST 20) or 40 (POST 40) sessions of PAS combined with exercise. MEPs and MVCs are significantly (p<0.05) increased after 20 sessions and significantly more increased after 40 sessions. (Bottom) Average (±SD) times to perform the Graded and Redefined Assessment of Strength, Sensibility and Prehension (GRASSP) (left), and the 10-m walk test (right) before and after 20 or 40 sessions and at a 9-month follow-up. The times are significantly decreased after 20 sessions and significantly (p<0.05) more decreased after 40 sessions, and the decreases persist at the follow-up 9 months after treatment ends. *: P<0.05. (From¹¹³.)

C. Brain-computer interface(BCI)-based feedback training of motor imagery improves hand/arm function in people with chronic stroke.

1. Statistical scalp maps (nose at top; stroke-affected hemisphere on the left) associated with tonic grasping movement imagery (MI) of the affected (left) and unaffected hands (right). The differences in EEG sensorimotor rhythm (SMR) desynchronization (decrease) (alpha and beta1 frequency ranges) between the (BCI) and control (CTRL) patient groups in the Pre-treatment (top row) and Post-treatment (bottom row) sessions were assessed by t test. Pixel color represents the corresponding probability value. Gray indicates nonsignificant differences; white–yellow indicates stronger desynchronization (p<0.05, Bonferroni corrected) in the BCI group; and black denotes stronger desynchronization (p<0.05, Bonferroni corrected) in the CTRL group. After treatment, SMR desynchronization is greater in the BCI group than in the Control group over the affected hemisphere and, to a lesser extent, over the other hemisphere.

2. Improvements in clinical outcome measures (Fugl–Meyer Assessment (FMA), Medical Research Council scale for muscle strength (MRC), National Institute of Health Stroke Scale (NIHSS)) in the BCI group and the Control (CTRL) group. *: p<0.05 between groups by independent-samples t-test. Improvement is greater in the BCI group by all three clinical measures; and the probability of achieving a minimal clinically important difference (MCID) for the FMA (7 points) was significantly (p=0.01) greater in the BCI group (11/14) than in the Control group (3/14). (From¹⁵⁵.)

Recent studies describe operant conditioning of the H-reflex in a specific phase of locomotion, and operant conditioning of the corticospinal motor evoked potential (MEP) and cutaneous reflexes; furthermore, they report therapeutic benefits in people with chronic stroke and multiple sclerosis.^143–148

Paired associative stimulation (PAS)

In people with chronic incomplete SCI, Perez and her research group are using noninvasive paired associative stimulation (PAS) to produce Hebbian plasticity at corticospinal-motoneuronal synaptic connections. Proof-of-principle studies suggested that it is possible to target spinal synapses using noninvasive electrophysiological methods in people with or without SCI.^149,150 Repeated paired pulses at precise intevals strengthen cortical control over spinal motoneurons.^{113,149,151–154} In their impressive series of controlled studies, they show that repeated sessions of PAS plus skill-specific practice produce clinically significant improvements in locomotion and in hand-arm function. Multi-site PAS produces greater functional improvements than single-site PAS.^113,154,243 The benefits last for at least 9 months after treatment ends. Most recently, Perez and her group have shown that a triple-strategy protocol combining generalized plasticity, targeted plasticity, and skill-specific practice further enhances recovery (Refs. ²⁴³ and ²⁰).

As with H-reflex conditioning, targeted plasticity at a critical site – in this case the corticospinal connection weakened by SCI – leads to much wider beneficial plasticity. Figure 6 B illustrates the physiological basis and functional effect of PAS combined with skill-specific practice. Hebbian plasticity increases corticospinal control, which enables better practice that improves heksor adaptations. Beneficial plasticity extends beyond the targeted plasticity. The widespread functional improvements persist after treatment ends, presumably maintained by ongoing heksor adaptations.

Enhancing electroencephalographic (EEG) correlates of sensorimotor function

In a controlled trial, Pichiorri et al. (2015) studied 28 people with impaired arm/hand function due to a stroke several months before.¹⁵⁵ Using noninvasive brain-computer interface (BCI)-based feedback, they targeted beneficial plasticity to the corticothalamic circuits that underlie the desynchronization (decrease) in electroencephalographic (EEG) sensorimotor rhythms (SMRs) preceding and accompanying active movement. Because greater SMR desynchronization precedes faster more accurate movement,¹⁵⁶ the BCI feedback sought to augment this desynchronization. When this targeted plasticity was combined with skill-specific practice, functional recovery was superior to that produced by skill-specific practice alone. Norman et al. (2018) described similarly positive findings in another group of people with stroke.¹⁵⁷

Figure 6 C illustrates the enhanced SMR desynchronization achieved by this BCI-based intervention, and summarizes the significant associated improvement in arm/hand function as measured by standard clinical tests. The probability of producing a minimal clinically important difference (MCID) in the Fugl-Meyer assessment (7 points) was significantly greater in the BCI group. As with PAS and H-reflex conditioning, combining targeted beneficial plasticity of a critical part of a damaged heksor with skill-specific practice led to overall improvement in function. A larger controlled study that seeks to confirm and extend these promising results is planned.¹⁵⁸ A variety of other recent studies further support the potential value of training EEG SMRs for enhancing recovery of motor function or enabling direct BCI-based control of an exoskeleton or other assistive device.^258–261

Dorsal root-specific stimulation during skill-specific practice

The previous sections describe noninvasive targeted-plasticity methods that improve function. In proof-of-principle invasive studies in people with chronic SCI, Courtine and his colleagues have implanted electrode arrays over posterior lumbosacral spinal cord and used them to provide precisely focused and timed epidural electrical stimulation (EES) of dorsal roots as a participant with SCI attempts body-weight supported locomotion (i.e., skill-specific practice).¹⁵⁹ Individual roots are stimulated at the times when they would be expected to provide sensory input to excite appropriate motoneuron populations.

Over five months of 4–5 sessions/wk training, lower limb function in the absence of EES improved substantially in the three participants; two became able to walk independently with crutches. It appears that the precisely timed posterior root stimulation produced strengthening of sensory afferent motoneuron excitation similar to the strengthening of corticospinal connections that Perez and her group produce with PAS.¹¹³ As with PAS, H-reflex conditioning, and BCI-based sensorimotor rhythm training, targeted plasticity at a critical site improves heksor adaptations and leads to wider beneficial plasticity and lasting functional improvement.

Skill-Specific Practice Plus the New Strategies: Combined Protocols

Skill-specific practice, the main strategy of present-day neurorehabilitation, gives damaged heksors the opportunity to make beneficial adaptations. Because the specifics of this practice vary across disorders, patients, and therapists, the heksor adaptations that result will also vary. For example, simply practicing a skill is likely to produce different heksor adaptations from training aimed at strengthening the muscles involved in the skill.^192,193 That said, skill-specific practice in some form will remain a central component of essentially all therapeutic protocols. Practice enables a damaged heksor to shape the available CNS plasticity so as to restore its key features.

Skill-specific practice is often effective, but seldom fully effective. Many disorders produce damage that cannot be repaired by the plasticity available in the mature CNS. Furthermore, practice may not take maximum advantage of the plasticity that is available. The two new strategies – generalized plasticity and targeted plasticity – address these deficits. At the same time, each of the new strategies has a potential risk.

The generalized-plasticity strategy increases the capacity for plasticity; it thereby increases the options that damaged heksors have for making beneficial adaptations. It provides them with new ways for repairing themselves; they can use the newly available plasticity to enhance recovery of their key features (e.g., Figures 4 & 5). At the same time, this strategy might possibly exacerbate the inability of skill-specific practice to ensure optimal use of the available plasticity. By offering even more possibilities for plasticity, the generalized-plasticity strategy might conceivably worsen this problem; it might even lead to maladaptive plasticity akin to that underlying dystonias that may afflict professional musicians and highly trained athletes.^{194,195,244–246} While animal studies seeking to demonstrate maladaptive plasticity have failed to do so, the possibility exists.¹⁹⁶ People with chronic damage that impairs sensory feedback may be particularly vulnerable (see below).^196,197 Neuropathic pain in the months post-stroke may reflect central (e.g., thalamic) abnormalities in sensory processing.^252–254 Maladaptive plasticity in only a few individuals might be missed when analyses focus only on group effects.

The targeted-plasticity strategy enhances a damaged heksor’s use of the available plasticity. Thus, it mitigates the problem caused by the lack of overall direction and may reduce the risk of maladaptive plasticity. At the same time, targeting beneficial plasticity requires knowing what to target. In some cases, such as Figure 6 A, this is clear; in others it may not be. Simply measuring function at a site will not always indicate whether the site should be targeted, or indicate what change would be beneficial (e.g., whether a specific synaptic input should be strengthened or weakened). A reflex of a given size may be beneficial in one person and harmful in another. Even if reflex function at a critical site is clearly abnormal, it may not be certain that this abnormality contributes to skill impairment.¹⁸⁷ Animal data indicate that incorrect targeting, while not beneficial, is not detrimental; the damaged heksor changes so as to prevent further impairment of its skill.¹⁹⁶ However, this may not always be true, particularly when the sensory feedback that guides heksor changes is impaired.¹⁹⁷

In sum, for both of the new strategies, preclinical and clinical studies must remain attentive to their possible risks. The most successful new protocols are likely to be those that combine skill-specific practice, the traditional strategy, with the two new strategies so as to maximize both their synergy and their ability to mitigate each other’s risks. While most studies of promising rehabilitation therapies have evaluated a single therapy, the value – and indeed the necessity – of therapeutic protocols that combine therapies is now recognized.^20,266,267

Selecting a promising new combined protocol to study is demanding. The many kinds and sites of possible plasticity, the many new methods for producing generalized or targeted plasticity, and the many kinds of skill-specific practice that might be combined with the new strategies generate many appealing protocols. Furthermore, once a protocol is selected, it must be parameterized: are the strategies combined in series or in parallel; how many trials; how many sessions; how often? Testing even one set of parameters is lengthy and expensive, especially in humans.

In silico models (i.e., computer-based computational models) can help.²⁴⁷ These models can simulate the neural processes underlying the functional impact of a proposed therapeutic intervention.¹⁹⁹ They can elucidate the mechanisms of recovery and predict functional outcomes for specific parameters, patient populations, or individual patients.(e.g.,^200,204,205) A model can predict in a few seconds the results of exposing thousands of patients to a novel therapeutic protocol for many months. Thus, these models can rapidly triage candidate protocols to identify the most promising ones for particular populations or even particular individuals. Furthermore, with similar rapidity, a model can predict the clinical impact of changing each model parameter over a range of possible values. It can thereby provide initial parameters for animal and human studies, and elucidate the difficult issue of therapeutic dosage (e.g.,^201–203). A model might even help to define a patient-specific treatment plan and guide appropriate adjustments in this plan as treatment progresses (e.g.,²⁴⁸). Modeling might also clarify the factors determining whether functional recovery results from repair of existing heksors or from creation of new heksors that produce skills with comparable key features.²⁰⁶

In studies particularly relevant to combined protocols that produce targeted plasticity, Reinkensmeyer and colleagues used a simple artificial neural network (ANN) to study the impact of skill-specific practice on recovery of function after a stroke.²⁰⁷ The model made predictions consistent with clinical findings and brain imaging data; it also suggested that skill-specific practice that targeted weaker nodes in the network (i.e., nodes that were more weakly connected to function) could enhance recovery. To test this, they used an advanced version of the same ANN to compare the ability of different protocols to restore finger extension after a stroke had destroyed many of the contralateral corticospinal neurons that normally produce extension.²⁰⁸

Figure 7 summarizes this study. It predicts the recovery of extension force produced by standard skill-specific practice and the recoveries produced by different combinations of standard and targeted skill-specific practice. It also predicts the associated changes in the contralateral/ipsilateral distribution of the corticospinal neuronal activity that produces extension. The ANN predicts that a combination of standard and targeted skill-specific practice will restore strength more than standard skill-specific practice alone, and will also restore a more normal topographical distribution of the corticospinal neurons that produce extension. Furthermore, the model predicts the particular mix of standard and targeted practice that maximizes recovery. The mechanism responsible for the superior result of the combination is clear. Standard practice does not optimize the performance of corticospinal neurons that are weaker and adapt more slowly. By focusing on those neurons, targeted practice does optimize them, and thereby increases finger extension. Such combined protocols might be translated into clinical practice.²⁰⁸

Figure 7. An Artificial Neural Network (ANN) predicts the efficacy of targeted plasticity after a stroke.

A: Network Architecture and Use. Left: Simulation of Standard Skill-Specific Practice. A two-layer feedforward ANN has n corticospinal (CS) neurons x_1-n with activation levels x_i. Each CS neuron excites the downstream motoneuronal pool (Σ) with a strength defined by a weighting factor w_i. In each movement practice trial, these activation levels are generated when the neurons receive the command to maximize finger extension force F. The spinal motoneuronal pool (Σ) then sums the pattern of weighted activations. A nonlinear function g_i implements the physiological observation that the contribution of any single CS neuron to the excitation of the motoneuronal pool saturates at some activation level. The ANN optimizes the activation pattern through reinforcement learning in consecutive practice trials in which extension force F is the teaching signal (i.e., the goal is to maximize F). Right: Simulation of Targeted Skill-Specific Practice. The ANN uses only a portion of the CS neurons available. This optimizes the activation levels of these particular CS neurons.

B: Force increase produced by movement trials before and after a contralateral stroke eliminates some of the strongest CS neurons. Before the stroke, force increases to near maximum over 20,000 trials. After the stroke, the maximum possible force (indicated as maximum recovery) achievable by the surviving CS neurons is lower than the pre-stroke maximum. After 20,000 trials of Standard Skill-Specific Practice, force remains far below maximum possible recovery. In contrast, after 20,000 trials comprising a mix of Standard and Targeted Skill-Specific Practice, force is substantially closer to maximum possible recovery. Solid lines show the mean results and shaded areas show the standard deviations of 10 simulations. Standard Skill-Specific Practice alone leaves substantial residual capacity for recovery. An appropriate mix of Standard and Targeted Skill-Specific Practice recovers much of this residual capacity; it achieves significantly greater force recovery (t=9.72, P<0.001).

C: Topography of task-related CS neuronal activation for left index finger extension after 20,000 movement trials. Left: Standard Skill-Specific Practice in the normal brain. The ANN optimizes primarily the activation levels of CS neurons in contralateral primary motor cortex; to a lesser extent, it also optimizes the activation levels of CS neurons in ipsilateral primary motor cortex. Center: Standard Skill-Specific Practice after a contralateral (i.e., right-hemisphere) stroke eliminates some of the strongest CS neurons. The ANN optimizes abnormal ipsilateral activation of CS neurons and diffuse activation of remaining contralateral primary motor cortex CS neurons. Right: An appropriate mix of Standard and Targeted Skill-Specific Practice. The ANN optimizes primarily the activation levels of CS neurons in contralateral secondary motor areas; to a lesser extent, it also optimizes the activation levels of CS neurons in ipsilateral primary motor cortex. Thus, it restores more normal laterality.

D: Recovery of residual capacity as a function of the percentage of movement trials that are Targeted Skill-Specific Practice trials. The solid line shows the mean and the shading indicates the standard deviation of 20 simulations. Recovery improves as the dosage of targeted trials increases up to 20%, and declines thereafter. (Full presentation and discussion:²⁰⁸.)

With further modifications, this simple ANN could evaluate other combinations of standard skill-specific practice with the generalized- and targeted-plasticity strategies. It might simulate the generalized-plasticity strategy by increasing the change in the magnitude weighting factor w_i that occurs with each learning trial (Fig. 7 A), or increasing the number of corticospinal neurons that can excite the motoneuron pool. The first method would increase the rate of improvement in Figure 7 B; the second would raise the Maximum Recovery line. The ANN might combine such generalized-plasticity therapy with standard and targeted practice. It could explore the impact of dosage and timing (e.g., blocked vs. distributed) of each strategy and the impact of their timing relative to each other (e.g., sequential vs. parallel). The ANN might also be modified to simulate specific disorders (e.g., stroke vs. SCI). Furthermore, an ANN can help to elucidate the cortical-subcortical interactions that underlie recovery after a stroke that impairs complex functions such as the control of individual fingers.²⁴⁹

Models can assist in exploring options in animal testing and at multiple stages in subsequent human testing and clinical translation of new protocols. Throughout this multistep process, progress can benefit from the primary strength of computer modeling – the ability to predict in a few seconds the efficacies of alternative protocols or of a given protocol with different sets of parameters. Modeling can simulate comprehensive initial evaluations of promising therapeutic protocols, evaluations that are often not practical in animals and rarely possible in humans.

Potential applications to cognitive/behavioral disorders

Up to the present, neurorehabilitation has focused on sensorimotor disorders. Nevertheless, the triple-strategy therapeutic framework (i.e., generalized plasticity, targeted plasticity, skill-specific practice) that follows from the heksor/negotiated equilibrium paradigm is likely to apply to cognitive/behavioral disorders as well. For example, recent evidence indicates that right and left amygdalae have different roles in fear processing.^256,257 Together with other data, this finding suggests that, in a person with post-traumatic stress disorder (PTSD) who is performing a task associated with anxiety, blockade of right amygdala might increase generalized plasticity, while concurrent stimulation of left amygdala might shape (i.e., target) that plasticity so as to reduce anxiety. In the future, these interventions might possibly be combined with operant conditioning or paired associative stimulation (PAS) therapies similar to those now applied to sensorimotor disorders (e.g., Fig. 6).

Validating New Protocols

Promising new protocols will usually undergo pre-clinical testing in animals. If these are encouraging, initial single-arm human studies are likely to follow. Initial small studies can parameterize and generate interest in a novel therapeutic protocol, and larger single-arm studies can establish safety and feasibility and give preliminary insight into efficacy. Properly formatted, they can also establish maximum tolerated dose for non-pharmacological as well as pharmacological therapies.²⁵⁵ If they include participants with a range of Functional Gait Assessment (FGA) level^209,210 and outcome measures across International Classification of Functioning, Disability and Health (ICF) domains,²¹¹ these first trials may also indicate the minimum level of function needed to engage in the protocol and also identify patient characteristics that are associated with and can predict a positive response to the protocol.²¹² Single-arm studies may also provide data for computational modeling that seeks to define optimal parameters (e.g., training intensity and duration) for subsequent controlled trials. If these initial studies establish the safety, feasibility, dosage, and probable efficacy of the protocol, larger phase-2 efficacy studies are likely to follow. The single most important requirement for these efficacy studies is that they include appropriate control arms.

Controlled trials are essential if a new protocol is to become an established therapeutic option. The results of single-arm studies – particularly positive results – are uninterpretable for three reasons. First, they may simply reflect a placebo effect. Second, they may result from nonspecific aspects of the protocol rather than from the intervention that is its unique attribute (e.g., reflex operant conditioning,⁶² paired-pulse facilitation¹¹³ or a specific training intensity and duration²¹³). Third, they may be inferior to results provided by current state-of-the-art rehabilitation in that they produce less improvement and/or have more frequent or severe negative side effects. Furthermore, single-arm studies are inefficient because they do not control for the substantial day-to-day and longer term functional variability common in people with chronic disabilities.²¹⁴ Single-arm studies with positive results can actually be counterproductive if they lead to clinical dissemination of therapies that are no better than – or even inferior to – standard state-of-the-art rehabilitation. Such therapies constitute what might be called “therapeutic noise” – therapies that distract clinician and patient attention, and provider resources from therapies that have been validated through controlled trials.

The control arms serve two important functions. First, they assess mechanism. An ideal mechanistic control is identical to the experimental arm except that it lacks the presumptive mechanism (e.g., reflex operant conditioning,⁶² paired pulse facilitation,¹¹³ paired associative stimulation,²¹⁵ vagal nerve stimulation¹⁰⁶). To the greatest extent possible, patient, investigator, and evaluator should be blinded as to which group the patient is in (e.g., through use of sham stimulation). Second, controls compare the new protocol to current state of the art therapy. This control might be provided by a third arm, or by incorporation of state-of-the-art therapy into both the experimental arm and the mechanistic control arm.

These phase-2 trials should gather additional data that are often not practical to acquire in subsequent larger phase-2 or phase-3 efficacy trials in which recruitment of sufficient numbers of patients is essential and often difficult. These additional data include biomarkers and other participant characteristics that may predict or correlate with their responses to the protocol. Subgroups of individuals based on differing physical, physiological, cognitive, psychosocial, and/or experiential phenotypes may differ significantly in their responses to therapy.^{198,216,217,251} Important pre-treatment data include anatomical and functional imaging studies, physiological measures (e.g., spinal reflexes, motor and somatosensory evoked potentials), quantitative and qualitative clinical assessments, and patient-reported functional status. Given that a high dose of skill-specific practice is usually needed for therapeutic efficacy, understanding what engages and motivates a person is important.^223,224,250 For people with stroke, feedback-mediated practice increases motivation and engagement more than practice without feedback.^225,251

The person’s activities (e.g., athletics) and life style are also relevant. Healthy people have a standard set of skilled behaviors, including locomotion, reach-and-grasp, and language; and these skills as performed appear similar across people. However, the heksors responsible for these skills are likely to differ substantially across individuals due to their differences in past activities. Athletic training, such as ballet, modifies how people walk.¹⁹⁰ The key features of locomotion are the same, but the responsible CNS activity may be significantly different. For example, highly trained dancers have depressed spinal proprioceptive reflexes.^189,218 Thus, their post-injury state and subsequent therapeutic response might differ from those of non-dancers. People who are highly trained in asymmetric sports (e.g., the discus thrower in Figure 3 A) may have locomotion heksors that are asymmetric at multiple levels from cortex to spinal cord.⁶¹ Thus, the locomotor impact of a right-hemisphere stroke might differ markedly from that of a left hemisphere stroke. Sanchez et al. (2023) identifies four different walking patterns in people after a stroke.²²¹ Similar exploration of walking patterns in healthy people might provide new insight into individual differences in the impact of a stroke. In people who are bilingual, the neural activity that produces their native language is considerably more complex than it is in people who speak only their native language.^61,219,220 Acquisition of the second language changes the heksor that produces the first. In short, pre-morbid differences in the heksors responsible for important skills are likely to affect the nature and severity of post-injury disabilities, and may underlie inter-individual differences in the efficacy of specific therapeutic protocols.

Campbell et al. (2000) stress the importance of using quantitative and qualitative methods to obtain a comprehensive assessment of trial results.²²² Outcome assessments should span the three ICF domains of Body structure/function (impairments), Activities (limitations), and Participation (restrictions).²¹¹ A hierarchy of assessments that spans the ICF domains may help identify responder and non-responder profiles and enable better patient-specific selection of therapeutic protocol.²¹¹ In addition to standard quantitative outcome measures (e.g., 10-meter & 6-min walk tests, box-and-blocks, functional gait assessment (FGA), Fugl-Meyer, Action Research Arm Test (ARAT)), measures of autonomic function and quality of life are important. Outcome assessment should also include the patient’s perspective regarding the experience and the value of the protocol. This is important for gauging the probable success of clinical translation. Ideally, these outcome measures, combined with knowledge of pre-morbid activities and life style can identify subgroups of individuals for which pre-morbid characteristics can predict therapeutic response and help guide treatment.

Outcome assessment is important before, during, and at the termination of therapy, and at later times extending out 6 months or more after the protocol ends. Selected assessments during therapy may track the onsets and courses of physiological and functional changes and the response to treatment. Learning when the experimental group deviates from the control group can help further define the dose-response relationship.²¹³ Dose may need to accommodate tolerance (e.g. fatigue), which is likely to vary across individuals. Individual differences in learning are well recognized, but the factors responsible are not well defined.²²⁶ Periodic assessment of selected functional measures over the course of treatment may provide new insights. The trajectory of change may vary widely from person to person.²²⁸ Body-worn sensors and/or sensorized objects can capture trial-to-trial metrics that reveal different performance strategies.^226,227 Ongoing assessments enable linear mixed effects regression to model changes in a primary outcome measure (e.g., walking, targeted reaching) as a function of group membership and time.²⁰² Such analyses provide important insights that are not possible when data are simply averaged across participants.

Most combined protocols now under study are noninvasive. New technologies such as wearable sensors and sensorized objects can enable patients to practice on their own at home. With further hardware and software development, many protocols should be amenable to home use with periodic remote oversight by a therapist. Given the wide range across patients in adherence to home-based therapeutic regimens, the evaluation of these regimens should incorporate as much as possible automated recording of the duration and quality of therapeutic engagement. Home-based practice could increase treatment dose, facilitate transfer of recovered skills to daily life, and make rehabilitation accessible to many more people. Furthermore, it could greatly reduce the cost of therapy and the difficulty of accommodating reimbursement limitations. Recent studies document the efficacy of technology-based home practice.^229,230 These new technologies may also assess therapeutic impact on the ICF Activities and Participation domains.

Conclusion

Neurorehabilitation is now one of the most active and rapidly growing areas of biomedical research. Full recovery of locomotion, speech, and other skills impaired by CNS injury or disease – thought impossible a few years ago – is now conceivable. Since the turn of the century, neurorehabilitation has focused on skill-specific practice. Now, new recognition of lifelong CNS plasticity and new understanding of the CNS substrate of a skill provide new therapeutic strategies; and new technologies broaden the range and augment the efficacy of protocols that combine these new strategies with skill-specific practice.

The CNS remains plastic through life. The substrate of a skill is a distributed network of neurons and synapses that may extend from cortex to spinal cord. This network has been given the name heksor (from the Greek hexis). Heksors change through life; each adapts as needed to maintain the key features of its skill, the attributes that make the skill satisfactory. For example, key features of locomotion include attributes such as upright posture, right/left symmetry, adequate balance, and acceptable metabolic cost. Muscle activity and kinematic details may change, but key features are maintained. These key features are the memory of the skill. The plasticity of the heksor ensures the stability of the memory. Heksors overlap each other; they share neurons and synapses. Through their concurrent adaptations, they keep the CNS in a negotiated equilibrium that enables each heksor to maintain its skill.

In light of this new understanding, the primary purpose of neurorehabilitation is to enable damaged heksors to repair themselves. This goal leads to two new therapeutic strategies. The first strategy increases plasticity in general. This provides a damaged heksor with more options for self-repair; these new options enhance the heksor’s adaptations during skill-specific practice. Promising methods include vagal nerve stimulation and agents such as NVG-291. The second strategy targets beneficial plasticity to a critical site in a damaged heksor. This improves skill-specific practice and guides the heksor’s adaptations; the result is much wider beneficial plasticity. Promising methods include paired associative stimulation of corticospinal connections and operant conditioning of a spinal reflex, motor evoked potential, or EEG rhythm.

These new strategies help damaged heksors to restore their skills and reestablish a satisfactory negotiated equilibrium. The most effective protocols are likely to combine both new strategies with skill-specific practice. Figure 8 illustrates the operation of such combined protocols. Determining which protocols work best for which populations, or for which patients in a single population is challenging. Controlled clinical trials and identification of predictive biomarkers are essential if the unprecedented therapeutic advances now possible are to be achieved.

Figure 8. 21^st-Century Neurorehabilitation.

A skill such as locomotion is produced by a heksor – a network of neurons and synapses (green Δs) that may extend from cortex to spinal cord (left image). A heksor continually adapts its neuronal and synaptic properties to keep its skill satisfactory. CNS injury (next image) may destroy (dashed Δs) or impair (red Δs) heksor neurons and synapses, thereby impairing the skill. Two new strategies – Generalized Plasticity (GP) and Targeted Plasticity (TP) (third image) – increase the efficacy of Skill-Specific Practice, the traditional strategy. GP therapies restore neurons or synapses (open Δs). After spinal cord injury (SCI), a GP therapy might enable transected descending axons to reconnect to spinal neurons. The heksor then shapes these restored connections through practice. TP therapies target beneficial plasticity to a critical site in a heksor (central green Δ). After SCI, a targeted plasticity therapy might reduce a hyperactive stretch reflex, thereby reducing clonus. This improves practice, which enables the heksor to achieve wider beneficial plasticity (final image). After practice, the restored heksor may differ from before injury (cross-hatch green Δs). Protocols that combine the new strategies with practice produce clinically significant improvements that persist.

Data Availability Statement

No new data are associated with this article.