Neurobiology of Rehabilitation

Cerebral Sensorimotor System

The cortex in humans contributes to reaching, grasping, individuated finger movements, and walking-related motor control. At least six motor projections, in addition to the dominant ones from the primary motor cortex (M1), excite the motor pools. Descending fibers from the primary sensory cortex (S1) also project within the corticospinal tract to the dorsal horns.³ Axons from M1 pass through the posterior limb of the internal capsule, dorsal and ventral premotor neurons project through the knee, and the supplementary motor area (SMA) fibers pass through the anterior limb of the internal capsule. Dorsal and ventral cingulate fibers are distributed among the latter regions. For example, the macaque's L-6–S-1 neurons, which contribute to hindlimb stepping, receive descending corticospinal tract projections from about 24,000 neurons in M1, 6000 from SMA, 6200 from dorsal and ventral cingulate, 5000 from dorsal premotor, and 10 from ventral premotor cortices.⁴ Each of these cortical regions interacts with visual, vestibular, aural, proprioceptive, cutaneous and other inputs to help plan, select, initiate, and maintain unilateral and bilateral skilled movements. Thus, the corticospinal tract, which includes some uncrossed fibers within the lateral and ventral funiculi, draws from neurons that are distributed and separated by somewhat different vascular territories. One assembly of neurons may partially compensate for loss of another when subjects find a strategy to activate spared neurons.

The neurons within M1 and other somatotopically organized sensorimotor regions are also mutable controllers of muscles and movements. Clusters of neurons connected by horizontal fibers alter their relative ability to represent a shoulder, wrist and finger movement depending on how much the practice of a skilled movement fires these neurons together.⁵ Electrical stimulation of clusters of neurons in M1 in trains lasting 500 ms reveals cells within adjacent sites that conduct rather stereotyped, but commonly employed movements, such as elbow flexion or extension that depends on the initial position of the arm, a hand-to-mouth pattern for feeding, and defensive postures.⁶ These homuncular organizations increase the flexibility of M1 in guiding complex actions. In addition, an injury that disrupts one assembly of neurons that participate in a movement may have nearby neurons come to represent aspects of that movement with motor skills retraining,⁷ although some of the improvement in behavior may arise from other portions of the cortical, sub-cortical, or spinal motor network.⁸

Rapid representational plasticity has been demonstrated using transcranial magnetic stimulation (TMS) of motor cortex by the practice of simple directional finger movements, which can be augmented by neuromodulators such as amphetamine.^9,10 One correlate of learning-induced plasticity is the synaptic expression of long-term potentiation (LTP) and long-term depression (LTD) in neocortex¹¹ as well as in the hippocampus. LTP is associated with the proliferation of dendritic spines.¹² This morphologic change has been found in homologous cortex opposite from the site of an experimental sensorimotor cortical lesion when the unaffected limb works to compensate for the paretic one.¹³

Corticostriatal neurons from primary and secondary motor areas are distinct from those within the corticospinal tract and respond especially to sensory inputs associated with the direction and force of movements. Uninjured cortical descending tracts from crossed and uncrossed projections may play a greater role during rehabilitation based on their inherent connectivity, perhaps especially if sensory feedback to the sensorimotor cortices is as typical of the desired movement as feasible. Cues from therapists to evoke cognitive strategies that bring these regions into greater play during training may increase the level of descending drive on motoneurons of the spinal cord. For example, ventral premotor neurons are most active during shaping and grasping the hand to hold an object. A visual cue, such as an object, may engage these neurons to enable a subject with a hemiparetic hand to reach and grasp, a movement that cannot otherwise be initiated when no object is present. Imagining a movement and watching a movement will activate many of the nodes in the sensorimotor network that are also active when a person carries out the actual movement.^14,15 Thus, visual practice could produce the cerebral reiterations that increase synaptic efficacy for learning a skill.

Brain Stem Nodes

The brain stem contains centers that contribute to the initiation of rhythmic flexion and extension for walking. The basal ganglia and cerebellum project to these locomotor regions. The dorsal mesencephalic locomotor center and the mesopontine locomotor center with its cholinergic and glutaminergic cells activate the lumbar spinal central pattern generators (CPGs) when stimulated electrically or with certain drugs. These brain stem regions project to reticulospinal nuclei that pass bilaterally into the ventral funciuli, providing another route for spared cortical and brain stem drives to activate motoneurons for movements. This pathway provides a slower and less precise control for flexor movements than the direct descending corticospinal tract. Several brain stem pathways have been shown in animal models to generate new dendrites after being damaged. For example, the corticorubrospinal fibers, which participate in distal more than proximal upper extremity movements for grasping, show spontaneous collateral sprouting¹⁶ from the intact hemisphere and functional reorganization¹⁷ that includes the corticospinal tract. Of interest, atrophic rubrospinal neurons can be coaxed with neurotrophins to regenerate axons after a chronic SCI.¹⁸ This pathway, with its connections to the cerebellum, can potentially substitute for corticospinal fibers for making fine hand movements.

The cerebellum monitors the outcome of every movement using proprioceptive inputs from the dorsal and ventral spinocerebellar tract. These inputs are also copied to the thalamus and motor cortex, as well as the brain stem locomotor centers. The timing of coordinated movement sequences, as well as computations on the position, velocity, acceleration, and inherent viscous forces of the moving limbs, is partly orchestrated by the cerebellar nuclei and Purkinje cells. The great interest in these afferent signals within cortical regions for motor control suggests that motor skills training during rehabilitation should aim to optimize kinematic and kinetic inputs that are associated with normal walking, reaching, grasping, and pinching.

Spinal Cord Systems

The spinal cord of humans probably includes CPGs for locomotor movements. These neural circuits produce oscillating patterns of flexion and extension, independent of sensory input or supraspinal commands. Elemental CPGs may control different muscles around each joint. These oscillators are interlocked by intrinsic connections and by their responses to segmental afferents and descending command centers.

Evidence for pattern generation comes from experiments in vertebrates, including non-human primates, in which the spinal cord is transected in the low thoracic region and deafferented from all dorsal root inputs below that level. Electrical stimulation and monamines placed on the isolated lumbar cord produce alternating electrical activity in the ventral roots of limb flexors and extensors.¹⁹ When cats and rats undergo spinal transection, their paraplegic hindlimbs lose the ability to step on a treadmill. With practice that emphasizes hindlimb loading and treadmill-induced hip extension, they regain alternating stepping movements, though the paws do not readily clear the surface. Noradrenergic agents may help initiate stepping and the training has lasting effects.^20,21 The cats and rats generally cannot step very well over ground with their hindlimbs, however. Animals trained to stand, rather than to step, walk poorly on the treadmill, pointing to the specificity of the type of practice on spinal cord learning.²² These findings suggest that a CPG is at work. Evidence for a CPG in humans has been found in both the spontaneous rhythmic movements made by some patients after SCI^23,24 and from the evolution of EMG activity in the legs of patients with paraplegia who are manually stepped on a treadmill²⁵ or undergo electrical stimulation of the dorsal horns at L-2.²⁶

Modulation of the CPG involves the organization of several neurotransmitters that seem to be conserved from lampreys to mammals^27–29 For example, glutaminergic reticulospinal neurons excite ipsilateral spinal motoneurons and interneurons and contralateral glycinergic inhibitory neurons. Glycinergic neurons also have axons that cross to the opposite half of the spinal cord CPG. Other amines and peptides modulate the initiation, maintenance, and termination of cell bursts. Metabotropic receptors for serotonin (5-HT), gamma-aminobutyric acid (GABA), and glutamate are also activated within the CPG network during locomotion. These neurotransmitters could be restored to the lumbar cord after a brain or SCI by systemic or intrathecal drugs, by implanted cells that release a neurotransmitter, or by regeneration of specific axons.

Other intrinsic systems may contribute to the flexibility of spinal control for reaching and for walking. A small set of modules appear to store components of flexor and extensor synergistic movements within the typical workspace of the extremities.³⁰ The modules may be activated in chains to achieve functional movements. This synaptic organization probably shares connections with the CPG and responds to segmental sensory inputs and descending controllers of the kinematics for reaching into space and walking.³¹ In addition, interconnections among columns of motor pools via propriospinal pathways aid postural adjustments and movement during reaching and ambulation, as do spinal reflex pathways and vestibulospinal inputs.

Sensory inputs provide a powerful source for functional modulation of the CPG, as well as for positive and negative force feedback during walking.³² In studies of higher vertebrates and human subjects, cutaneous inputs from the sole and Ia and Ib inputs to the hips and ankles are especially important drives for walking.^32,33 For example, the spinal cord responds to varying levels of weight bearing on the legs in patients with clinically complete SCI, reflected in changes in the amplitude and timing of electromyographic bursts from lower extremity muscles.³⁴ Studies of cats and humans reveal the impact of the timing of hip extension at the end of stance of one leg and simultaneous loading of the opposite leg for successful initiation of the swing phase.^35,36 These sensory inputs presumably contribute to the internal models the brain possesses about the properties of limbs and limb mechanics.³⁷

Motor Control

Theories about motor control and the acquisition and recall of motor skills are beginning to play an important role in the development of more sophisticated rehabilitation strategies. M1 is involved in the initial phase of learning a motor skill, as well as in early consolidation from an unstable to a stable state.⁴⁹ Lasting learning of a simple, but novel motor skill in non-human primate studies requires a considerable number of practice repetitions, from 300–1500, to reveal behavioral and neuronal reorganization changes.^50,51 After a brain or spinal cord lesion, the nervous system has less information about how to select, initiate, and correct movements, and even the mechanical properties of the joints and muscles may change, so both attempted actions and the process for relearning functional movements may suffer. Even greater intensity and duration of practice become necessary.

Many parameters for an internal model of interactions with the environment have been investigated. At the neuronal level, for example, firing rates during learning to reach to a target at a specific angle within the body's workspace increase within the subpopulation of cells that are preferentially tuned to the direction of the target. This activity seems most related to a modification in the internal model of movement kinematics for computation of a visuomotor transformation, rather than to movement dynamics.⁵² One especially relevant theory of motor control suggests that enough feedback control for movement can be obtained from optimal estimates of the state of a limb, using parameters such as joint angles or muscle lengths.^37,53,54 The system controls the global goal of a task from low-level signals, each concerned with a portion of the system. Errors that influence motor performance are corrected and signals that are not relevant are ignored. This theory suggests that spared neural nodes may be able to act as controllers by optimally selecting afferent feedback.

Another theory suggests that neural signals may explicitly encode the endpoint of the limb, such as the cat's paw during walking, and that the dorsal spinocerebellar tract provides this kinematic information.³¹ Translating models built upon studies in cats with their bi-articular hindlimb muscles into humans may be misleading, however, since these muscles may naturally reflect end-points more than joint angles. Another point of view is that the brain may issue motor commands based on a prediction of the forces for an upcoming movement. An internal model of experienced forces also generalizes to upper extremity parameters such as velocity and position in space.^55,56 Activation of the N-methyl-D-aspartate (NMDA) receptor and inhibition by GABA were shown to be involved in the acquisition, but not the recall of a new internal model of the dynamics for reaching.⁵⁷ These theories are important to the neurobiology of rehabilitation, because they help set the tone for styles of practice, the sensorimotor parameters to be monitored to optimize training, the neural pathways that need to be engaged for skills learning, and the potential for pharmacologic interventions to augment motor learning.

Massed Practice of Task-Oriented Motor Skills

The essence of therapy for any disability is practice. A practice session can have a powerful, but only temporary effect. A positive effect on performance during a training session by repeatedly practicing the same movement may not lead to long-term learning. Studies of interventions should include a doseresponse curve to establish how much practice is needed to achieve a retraining goal. During practice, contextual interference from intermixing other related tasks may enhance learning, unless cognitive impairment impedes attention or procedural learning.

Treadmill Training

Body weight-supported treadmill training (BWSTT) was derived from the treadmill training approach for cats after complete spinal cord transection in studies of CPG activity. BWSTT, in theory, allows the spinal cord and supraspinal locomotor regions to experience sensory inputs that are more like ordinary stepping compared to the atypical locomotor inputs created by compensatory gait deviations and difficulty loading a paretic limb.^58,59 More typical proprioceptive and cutaneous input, as noted earlier, may improve the timing and increase the activation of residual descending locomotor outputs on the motor pools. Most important, BWSTT allows massed practice at different walking speeds and levels of limb loading with repetitions guided by the cues of the therapist. Randomized clinical trials for patients with hemiparetic stroke have produced mixed results,^60,61 but treadmill speeds have not been optimized.⁴⁴ A multi-center trial of patients with acute incomplete SCI tried to optimize training at high treadmill and overground walking speeds for 12 weeks with best-of-possible kinematics and kinetics, but demonstrated no significant differences from the conventionally trained subjects.⁶² The trial established a reproducible retraining approach that can serve as an experimental control for the style and intensity of locomotor rehabilitation in future clinical trials of pharmacologic and biologic interventions.

Constraint-Induced Movement Therapy (CIMT)

Rehabilitation practice-induced neuroplasticity and behavioral gains have been repeatedly demonstrated for the upper extremity in patients who retain at least modest motor control.⁶³ In the most common paradigm, subjects practice with a therapist for at least six hours a day for two weeks on a variety of tasks with the affected arm plus restraint of the normal hand for most of the day. These patients can, at onset, dorsiflex the wrist at least 10 degrees and partially extend the fingers of the paretic hand.^64,65 Less intensity may work as well.⁶⁶ The most important aspect of this approach is massed practice and feedback about movement skills that are important to the subject, rather than the type of restraint. Animal models of forced use early after an ablative or traumatic focal cortical injury suggest an increase in the volume of the lesion and behavioral deficits, probably on the basis of glutaminergic toxicity.⁶⁷ The intensity of use of the limb, however, was far greater than any clinical situation could allow. Forced nonuse of the limb affected by experimental damage restricted to the striatonigral dopamine projections, in contrast, augmented dopamine loss and Parkinsonian symptoms.⁶⁸ One acute clinical trial of the approach showed positive results⁶⁴ and data from a multicenter trial for hemiparetic patients who are 3–9 months post-stroke are pending.⁶⁹

Biofeedback and Automated Robot-Assisted Devices

Biofeedback (BFB) includes a variety of instrumented techniques that try to make the treated subject aware of physiologic information that can be used to better train an activity. Electromyographic BFB to increase the amplitude of muscle contractile bursts, decrease co-contraction of muscles, improve the timing of a contraction, can enhance skilled movements.⁷⁰ Robotic devices aim to maximize practice with only intermittent therapist oversight.¹ One robotic exoskeleton manipulates a patient's paretic elbow and shoulder by a two-degrees-of-freedom impedance controller system, much as a therapist might provide hand-over-hand therapy for reaching in a plane across a table. Motor power and control improved at the shoulder and elbow with this form of robotic training, consistent with the greater intensity of practice with those muscle groups.⁷¹ Active participation improves function more than passive movement, as might be expected during motor learning.⁷² By providing data on the intensity, duration, and accuracy of practice, these devices allow future studies of parallels between therapy-induced behavioral gains and activitydependent reorganization, perhaps monitored by functional neuroimaging techniques.

Pharmacologic Augmentation

Any neurobiology of rehabilitation must consider the potential to augment training strategies with medications that act on neurotransmitters, neuromodulators, and intracellular second messengers. The goals include strengthening synaptic efficacy within perilesional neurons and other nodes of the motor network during task learning, replenishing neurotransmitter projections that have been disconnected to reverse diaschisis, and preventing transsynaptic degeneration of neurons.

Among these drugs, dextroamphetamine increases the cortical signal-tonoise ratio.⁷³ Cholinergic projections serve as a gate for behaviorally relevant sensory information.⁷⁴ Human and animal studies have provided preliminary evidence that a variety of medications, such as dopaminergic and noradrenergic,^75–78 cholinergic,⁷⁹ and serotonergic⁸⁰ agents, may facilitate the rate or degree of motor recovery. Drugs, along with other neurostimulatory approaches, may augment gains in slow learners more than in subjects who can quickly learn a skill.^73,81 Drugs may also activate or inhibit subcomponents of the distributed sensorimotor system, such as the CPG.²¹ Blockers of dopamine and norepinephrine may inhibit skilled motor gains,⁸² perhaps depending on the time of use in relation to the injury.⁸³ A few studies suggest that intensive speech therapy combined with a drug that enhances vigilance or learning may benefit patients who have adequate language comprehension.^84,85 The rapid growth in knowledge about the molecules that modulate learning, such as agonists of the NMDA receptor, cyclic nucleotide adenosine monophosphate (cAMP), and cAMP response element binding protein (CREB), is leading to the possibility of new lines of drugs to augment rehabilitation strategies.^86,87

Controlled trials of anti-spasticity agents have varied widely in the target symptoms managed and the outcome assessments employed.^88–91 Functional gains related to walking and use of the upper limbs are often marginal. However, a medication that prevents disabling spasms may improve quality of life. Continued basic studies of the neurobiology of spasticity, such as the windup of flexion reflexes and other physiologic changes induced in the cord by loss of supraspinal input, are needed.^92–94

Randomized trials that compare a rehabilitation intervention combined with an experimental drug versus a placebo require considerable thought. For example, both the dose of medication and the dose of the rehabilitation strategy need to be optimized and adverse effects need to be minimized. Outcome measures should be sensitive to important changes in function and relevant to the intervention. The choice of pharmacologic augmenting agent, at least for sensorimotor studies, may be developed from TMS, positron emission tomography (PET), and functional magnetic resonance imaging (fMRI) studies that reveal a drug-induced increase in cortical excitability, rapidly induced plasticity in M1, or change in neurotransmitter levels in patients.^95–97

Neurostimulators and Neuroprostheses

Regional electrical stimulation of the cortex, deep nuclei, spinal cord, and motor unit could augment retraining. Phasic electrical stimulation of nervemuscle can stimulate genes to increase muscle fiber volume. When optimal afferent stimulation parameters are employed, peripheral electrical stimulation can augment cortical excitability and reorganization to enhance motor skills,⁹⁸ especially if coordinated with retraining. Deep brain and vagal nerve electrical stimulators and subdural or implanted cortical arrays may drive excitatory and inhibitory outflow to the forebrain and brain stem.⁹⁹ In theory, finding the optimal parameters for stimulation could modulate attentional drives and frontal lobe executive functions¹⁰⁰ and augment the acquisition of procedural or declarative learning. Repetitive TMS is another potential tool to excite or inhibit cortical pathways for sensorimotor⁸¹ or affective and cognitive processes such as hemi-attention.^101,102

Neuroprostheses have used local cortical potentials to control an electrical stimulator for anticipated movements.¹⁰³ Brain-computer interfaces that employ a variety of brain signals to communicate or to control a prosthesis without muscle stimulation¹⁰⁴ and neuroelectronic chips implanted into the brain to make a circuit¹⁰⁵ may both add to our understanding of the neurobiology of the brain and enhance functional outcomes for highly impaired patients.

Relevance of Animal Models

Much of the neurobiology of rehabilitation is drawn from animal models of injury and repair.¹ The translation of these experiments into rehabilitation interventions is not likely to proceed without discouraging setbacks, if the experience with acute neuroprotective interventions in patients with stroke, SCI and cerebral trauma holds.^139–142

Type of induced injury, timing of the intervention, location of lesion, relative volume, natural history of recovery, co-morbid conditions, sex and age, levels of activity, and other factors are controlled in laboratory experiments, but not in patients.^{140,142–145} Laboratory animals are bred and maintained in relatively unchallenging, impoverished and stressful circumstances, which may make their biologic responses to an injury different from wild rodents and humans.^146,147 Highly inbred rodent strains and transgenic mice allow the study of particular processes of injury and repair, but the cascades of gene expression over time and cellular and molecular changes in the milieu may not unfold in another strain or species, or in humans. Even sensorimotor, locomotor, and cognitive abilities vary between laboratory animal strains,¹⁴⁸ which may confound outcome measurements.

The pathways taken by neural cells and axons during development span distances of a few mm under the outer surface of the neural tube. The paths to targets are both short and sweet—multiple guidance signals that constrain and beckon appear in an orderly sequence within a highly organized, canallike matrix of capillaries and glia. For neural repair in adults, the distances that cells may have to migrate or axons regenerate differ dramatically between humans and rodents. The surface area of a mouse brain is 1/1000th that of the human brain. About 20 cross-sections of the rat lumbar cord can be superimposed within the cross-sectional area of the human cord.¹³⁴ Regenerating axons and collateral sprouts in experiments usually extend only 10–15 mm below a spinal cord lesion. Successful biologic interventions in animal studies reveal only modest numbers of short-lived, regenerating cells in models of stroke¹¹⁸ and modest numbers of regenerating axons after manipulations for SCI.^149–152 As with transplanted cells, the axons must operate within an environment that lacks the ideal ratio of signaling substances and targets that made survival, migration, and functional connectivity over tiny distances an evolutionary wonder that is the study of developmental neurobiology.

Still, if 10% of a supraspinal pathway can be restored,^153,154 then aumented by collateral sprouting, enough connections for rebuilding simple skills may be in place, even if the new inputs only reach propriospinal pathways below a spinal cord injury. Rehabilitation strategies that make use of the neurobiology underlying skills learning within a flexible, distributed motor system can then incorporate newly connected nodes into the motor controllers that lessen the disability of patients.