Neural plasticity and its contribution to functional recovery

DEFINITION

The idea that the cerebral cortex is dynamically organized was proposed in 1912, when Brown and Sherrington stimulated the motor cortex of chimpanzees and found that “a point which began by yielding primary extension may come to yield primary flexion in the latter part of the stimulation series” (Brown and Sherrington, 1912). In many investigations since then these phenomena have been referred to as neural plasticity. Neural plasticity can be defined as the ability of the central nervous system (CNS) to adapt in response to changes in the environment or lesions. This property of the CNS may involve modifications in overall cognitive strategies to successfully cope with new challenges (i.e., attention, behavioral compensation) (Bury and Jones, 2002), recruitment of new/different neural networks (Johansen-Berg et al., 2002; Fridman et al., 2004; Lotze et al., 2006; Heuninckx et al., 2008), or changes in strength of such connections or specific brain areas in charge of carrying out a particular task (i.e., movement, language, vision, hearing) (Cohen et al., 1997; Grefkes et al., 2008). At the cellular level, changes in membrane excitability, synaptic plasticity, as well as structural changes in dendritic and axonal anatomy as measured in vivo and in vitro may be demonstrated in animals and humans (Clarkson et al., 2010; Li et al., 2010). The study of neuroplasticity engages scientists from many different disciplines because of the profound implications it has for understanding the functional underpinnings of action and cognition in the healthy and lesioned brain (Dimyan and Cohen, 2010). Mechanistic understanding of neuroplastic changes in the process of functional recovery following brain lesions, one of the focuses of this volume, is already starting to lead to the development of more rational strategies to facilitate neurorehabilitation (Taub et al., 2002; Cheeran et al., 2009).

At a cellular level, neuronal circuits consist of synaptic connections between axons and dendrites. As these circuits extend over the brain there is the potential for a large number of possible interactive combinations allowing for great flexibility. Modification of sensory input may induce rapid changes in cortical representations through various mechanisms including unmasking of connections that are silent in the native state (Calford and Tweedale, 1991a, b). For example, blocking inhibition pharmacologically within a small region of the primary motor cortex (M1) immediately unveils new representational patterns (Jacobs and Donoghue, 1991), through unmasking horizontal excitatory connections previously hidden by inhibitory neurons. The strength of these horizontal connections and the balance of excitation and inhibition appear to shape cortical representations. Corticofugal connections make extensive long-range (±1 mm) links with other pyramidal tract neurons, and with local inhibitory interneurons (Landry et al., 1990; McGuire et al., 1991). It is now known that long-term potentiation (LTP) can be induced in these horizontal connections of adult M1, contributing to long-lasting associations among neurons within a motor cortical area (Hess and Donoghue, 1994). Moreover, vertical synaptic pathways in M1 can experience short-term depression, short-term facilitation, long-term depression and, under conditions of disinhibition, also LTP (Castro-Alamancos et al., 1995). In addition, slower, progressive plastic changes can be driven by learning (Robertson and Irvine, 1989; Chino et al., 1997), competition with other inputs (Merzenich et al., 1983), and use (Nudo et al., 1996b).

Basic science investigations have substantially advanced our understanding of the mechanisms of plasticity and metaplasticity, important in multiple areas of human cognition such as learning and memory, and in functional recovery from lesions in the CNS, as in stroke (Buonomano and Merzenich, 1998; Floel and Cohen, 2006). The term “metaplasticity” is often, but incorrectly, used interchangeably with “homeostatic” plasticity (see below) (Abraham and Bear, 1996; Fischer et al., 1997; Gentner et al., 2008; Jung and Ziemann, 2009). In the past few years it has become evident that these findings have direct implications for the way in which human disease is treated, and new efforts have been invested in research that translates these advances in the basic science domain to the formulation of new, rational strategies for promoting recovery of function in humans. To accomplish this goal, it is important to demonstrate that similar principles to those described in animal models apply to the human cerebral cortex in relevant behavioral settings.

SITESOF PLASTICITY

In most cases, the cerebral cortex has been the target of studies of human plasticity (Wolpaw and Tennissen, 2001). However, reorganization requires fine-tuning of activity at cortical as well as subcortical sites. In the motor domain, for example, spinal processes play a role in modulating locomotor learning (Bizzi et al., 2000) and plasticity after amputations and nerve transections (Wu and Kaas, 1999). Plastic changes following deafferentation can be identified at cortical (Kaas et al., 1983) and subcortical (Devor and Wall, 1981) sites. The extent to which plastic changes detected at cortical levels reflect reorganization in subcortical structures is incompletely understood and still underinvestigated (Wu and Kaas, 2000). Therefore, it is important to keep in mind that the neural substrates of recovery of function are likely distributed over multiple sites at different levels of the neuroaxis and not restricted to one specific location. It still represents a challenge to understand how these different levels interact with one another to accomplish a particular behavioral goal.

WINDOWOFOPPORTUNITY

Neural plasticity occurs throughout the life span (Elias and Wagster, 2007). During normal human development the CNS must continue to optimize performance and learn and adapt in the presence of changes in anatomical constraints (such as, for example, changes in limb length or muscle mass or strength) and experience (Gaillard et al., 2000). Additionally, neuroplastic changes identified following CNS abnormalities during development have been particularly impressive given their ability to reestablish almost normal behavior (Chen et al., 2002). One such example is the substantial recovery of motor function or language in children posthemispherectomy, implemented to ameliorate intractable seizures (Vargha-Khadem et al., 1997). The potential of neuroplastic changes to influence behavior and recovery of function was first widely accepted in relation to the developing brain. Only more recently was it understood that neuroplastic changes of substantial clinical relevance could occur in the adult CNS and in the elderly (Merzenich et al., 1996). It has now been proposed, for example, that recruitment of wider brain networks in the elderly and after stroke may play a beneficial role in maintaining the ability of individuals to carry out specific tasks or even in facilitating relearning (Heuninckx et al., 2008; Hummel et al., 2010).

FUNCTIONAL RELEVANCE

Plasticity of cortical representations within and across different brain regions is thought to represent the neural basis underlying sensory substitution, for example in blind and deaf humans (Rauschecker, 1995), as well as in the recovery of motor function after cortical lesions like stroke (Nudo et al., 1996a). Although neuroplasticity, as defined above, is a ubiquitous phenomenon (our brain is constantly changing), it may have different impact on different behaviors. It may be beneficial (often referred to as adaptive plasticity, the most common forms of plasticity studied; Cohen et al., 1997; Lee, 2009), have no influence (representing only epiphenomena of the modified behavior), or even result in deleterious consequences (i.e., maladaptive; Flor et al., 2006) on performance of particular tasks or sensory experiences. This concept has been referred to as functional relevance of neuroplasticity. Conceptually, it would not be surprising that plastic changes in, for example neuronal networks, may have beneficial implications on a particular behavior but at a cost to other behavior (Chklovskii et al., 2004). This concept of cost of neuroplastic changes, which has been to some extent overlooked, is starting to receive attention. Understanding these changes and how they can be influenced is pivotal in developing better treatments and therapies for patients (Hodics et al., 2006; Hummel and Cohen, 2006; Cramer, 2008).

PLASTICITY, METAPLASTICITY, ANDHOMEOSTATIC PLASTICITY

Plasticity likely depends on multiple mechanisms evolving on different temporal scales – minutes to months, even years. Rapid onset-mechanisms, which may operate over a limited period of time, are believed to represent initial steps of more slowly evolving processes of reorganization through which functional gains (or losses) may be sustained (Classen et al., 1998; Kleim and Jones, 2008). At the level of neuronal synapses, multiple transformations may occur from relatively short-lasting LTP, which appears to be largely independent of protein synthesis, to long-lasting LTP, which may persist along the life span. These synaptic changes are complemented by changes in neuronal excitability and structural changes, with the latter ones being detectable using light microscopy, for example. From a behavioural perspective, consolidation refers to a process that results in enduring performance improvements (Cohen et al., 2005; Krakauer, 2009) that is underway during training or learning but typically occurs after. Through consolidation, newly acquired skills become more robust in the face of disruptive experiences or may even improve further, a process termed off-line learning. Through reconsolidation, stored memories may be purposefully modified in order to strengthen or weaken them (Censor et al., 2010).

Metaplasticity refers to the influence that baseline neural activity immediately preceding presentation of a plasticity-inducing protocol (for example in vitro theta burst stimulation) can substantially influence the ability of neuronal elements to exhibit plasticity. The functional significance of metaplasticity may include, but is not limited to, controlling homeostasis of neural network excitability, for example, by virtue of modifying synaptic efficacy in an operational range. Metaplasticity may, therefore, protect against potentially noxious excitability increases (Abraham and Bear, 1996). It should be kept in mind that the specific characteristics of the baseline activity preceding application of the same plasticity-inducing protocol can then result in opposite effects on synaptic efficacy (Seol et al., 2007). More attention is now paid to these phenomena in humans because it is thought that they can substantially impact information coding and cortical reorganization (Gentner et al., 2008; Jung and Ziemann, 2009).

An example in humans is that the magnitude of the response to noninvasive brain stimulation protocols applied to the primary motor cortex critically depends on the previous history of neural activity (Ziemann et al., 1998; Iyer et al., 2003; Gentner et al., 2008). As long as the modulation is confined to the magnitude, but not the sign, of responses (increases vs. decreases in excitability, for example), most of these findings may be interpreted within the framework of homeostatic plasticity as proposed in the Bienenstock–Cooper–Munro (BCM) theory of synaptic plasticity (Bienenstock et al., 1982; Abraham and Bear, 1996). Fundamental to the BCM theory is a time-variable induction threshold. For example, prolonged low levels of postsynaptic activity decrease the induction threshold, thereby increasing the probability for LTP. Alternatively, a history of enhanced postsynaptic activity would increase the threshold for LTP and therefore increase the likelihood for long-term depression induction (Ragert et al., 2009). Based on this combination of basic science and human neurophysiological evidence, it is attractive to speculate that the response to motor training protocols could depend on the history of activity at the time training is imparted. In other words, activities carried out in the period of time preceding the actual rehabilitative treatment (sleep, caffeine intake, reading, feeding, etc.) could have substantial influence, so far not well characterized, on outcomes and perhaps to some extent contribute to well-described interindividual variability in treatment response. Human studies have indeed provided experimental support for a homeostatic model of plasticity (Jung and Ziemann, 2009). On the other hand, experimental or therapeutic manipulations applied after the treatment may also provide an opportunity for modulating the ultimate behavioral response (Reis et al., 2009).

GENETICINFLUENCES

Genetic factors can influence the human brain’s ability to experience neuroplastic changes. For example, a genetic polymorphism (Val66Met) in brain-derived neurotrophic factor (BDNF) reduces electrophysiological measurements of training-dependent plasticity of the primary motor cortex (Kleim et al., 2006; Cheeran et al., 2008). Although the implications for motor learning and recovery of function need to be firmly established, these factors could partially explain interindividual variability in functional recovery or response to pharmacological or training-based interventions. BDNF is almost certainly one of the first of many possible genetic polymorphisms that affect training-dependent plasticity and also the ability to learn (Fritsch et al., 2010). Other mechanisms such as polymorphisms in the gene coding for catechol-O-methyltransferase, serotonin transporter, and other proteins involved in modulating or regulating neurotransmission are being explored and will likely lead to more individually tailored rehabilitative protocols after brain lesions (Pearson-Fuhrhop et al., 2009).

NONINVASIVETECHNIQUESCAPABLE OF EVALUATINGNEUROPLASTICITY IN HUMANS

One important factor that contributed to substantial advances in the understanding of neuroplasticity at a systems level in the human brain has been the development of techniques that allowed the noninvasive measurement of these changes. Techniques like positron emission tomography, magnetic resonance imaging (MRI), both functional (fMRI) and structural (particularly diffusion tensor imaging, DTI), magneto (MEG) and electro (EEG) encephalography, and transcranial magnetic (TMS) and direct current stimulation have all played important roles in the noninvasive evaluation of neuroplastic processes associated with recovery of function after CNS lesions. These techniques provide information on the possible relation between anatomical connectivity (DTI) or functional activity in specific brain areas as well as interactions between neural networks (fMRI) and a particular behavior, recovery process, or response to treatment. fMRI has excellent spatial resolution but less sharp temporal resolution and alone does not allow firm conclusions on cause—effect links between these associations (Cohen et al., 1997).

Dramatic increases in readily available computational power have led to the development of novel analytical approaches to functional imaging. Some of these apply economic theories, such as structural equation modeling (Simon, 1953) and Granger’s causality (Granger, 1969), to model the interactions between (sub)cortical regions (Deshpande et al., 2009; Kim and Horwitz, 2009). Dynamic causal modeling also explores regional interactions but does so within a Bayesian framework (Penny et al., 2004). These approaches explore interactions that are overlooked by conventional activation analysis (Rowe et al., 2002).

Model “free” analysis using multivariate statistical approaches (such as independent component analysis and principal component analysis) have established the existence of resting state networks (De Luca et al., 2006). The study of these networks represents a fertile area of research given their ability to experience neuroplastic changes, for example in relation to learning (Albert et al., 2009; Mantini et al., 2009). Changes in resting networks may be lost in classical fMRI study designs that involve contrasting one condition (task) with a control condition (most commonly rest). Another development that takes advantage of the increasing sophistication in these analytical tools is real-time fMRI (deCharms, 2008). The excitement of this approach is that it potentially can demonstrate the ability of training in human and nonhuman subjects to recruit specific brain regions or neural networks.

Although a lot of work has focused on the evaluation of dynamic changes in neural networks, there is mounting evidence that motor training can induce structural changes as well. These include changes in gray matter density (measured using voxel-based morphometry) (Smith et al., 2007; Wrigley et al., 2009) and in white matter (measured using DTI) (Johansen-Berg, 2007; Ciccarelli et al., 2008; Johansen-Berg and Rushworth, 2009; Scholz et al., 2009). Structural changes in gray and white matter have also been described in the elderly (Boyke et al., 2008). A word of caution: the cellular changes that underpin these changes are not yet clear. Nevertheless, in healthy volunteers, learning to juggle produces changes in gray matter density as well as in white matter (Scholz et al., 2009) in biologically plausible regions. Although there are numerous technical difficulties that have prevented the application of these techniques to the lesioned brain, some interesting studies are starting to appear in patient populations. In a recent study it was shown that constraint-induced therapy, a treatment proposed to improve motor function after stroke (Wolf et al., 2006), induced increases in matter density in the affected hemisphere, which is in keeping with functional MRI data (Gauthier et al., 2008). There were also increases in gray matter density in the nonaffected hemisphere, which are difficult to predict in our current framework of understanding. In some cases, identification of these changes allows the formulation of predicting algorithms after stroke (Stinear, 2010).

In contrast to MRI techniques, MEG and EEG allow a millisecond by millisecond analysis of the activity in functional networks in relation to behavior (Birbaumer and Cohen, 2007; Pantev et al., 2009). As such, they can provide information on the timing of neuroplastic changes or serial processing in a way that imaging techniques alone still cannot. Additionally, MEG and EEG do provide important information on activity in neural networks with a very accurate temporal resolution (Fujioka et al., 2006). Of note is that activity originating in specific brain networks as measured with MEG has been successfully used to control a hand orthosis that controls movements of a completely paralyzed hand after stroke (Buch et al., 2008). Similar results in terms of output control using EEG in completely paralyzed patients have been reported previously (Birbaumer et al., 1999; Wolpaw and McFarland, 2004).

Noninvasive brain stimulation techniques have contributed in different ways to the evaluation of systems’ neuroplasticity in the healthy and lesioned brain. In particular, TMS allows the evaluation of the behavioral consequences of disruption of activity (virtual lesion) in relatively focal brain regions, for example those shown to be active during a particular behavior in fMRI studies (Pascual-Leone et al., 2000). Disruption of a specific behavior as a consequence of TMS-induced disruption of a particular brain region has been interpreted as indicative of a cause—effect link between the two (Cohen et al., 1997). In this sense, neuroimaging and TMS virtual lesion experiments are complementary. An example of the way in which these two techniques operate to address fundamental hypotheses in motor control has been the evaluation of the role of the supplementary motor area or primary motor cortex in motor learning (Muellbacher et al., 2002; Perez et al., 2007; Censor et al., 2010). In another example of how these tools can be creatively utilized, TMS has been applied in the fMRI environment to examine fundamental questions of interregional connectivity within neural networks, otherwise impossible to address experimentally (Bestmann et al., 2005). This armamentarium has created an important momentum in human systems neuroscience, making possible the experimental evaluation of hypotheses until recently beyond the scope of investigation.

MODULATIONOF NEUROPLASTICITY

Given its proposed influence on learning processes and recovery of function, one goal of present investigations has been to develop strategies to modulate neuroplasticity: to facilitate it when it plays an adaptive function and downregulate it when it is maladaptive (see above). Different approaches have been tried in animal and human settings. Use of pharmacological agents like amphetamine, L-dopa, or selective serotonin reuptake inhibitors (SSRIs) (for review see Floel and Cohen, 2010) in association with motor training protocols may result in behavioral gains accompanied by cortical reorganization in humans (Tardy et al., 2006). Clearly, the development of better training protocols that take into account advances in basic science is an important area of research (Luft et al., 2004; Wolf et al., 2006; Ramachandran and Altschuler, 2009). Examples of proposed new rehabilitative paradigms include the combination of customarily used training protocols with action observation (Stefan et al., 2005), motor imagery (Sharma et al., 2006; Page et al., 2009; Sharma et al., 2009), and focused attention (Stefan et al., 2004) thought to ameliorate function by facilitating the mirror neuron system (Nelissen et al., 2005).

In addition to these approaches, recent years have seen the formulation of interventions based on better understood principles of intracortical interactive functions (see Fig. 1.1 for a summary of the interregional influences on M1). For example, the identification of persistent interhemispheric inhibitory interactions between the primary motor cortices after stroke (Murase et al., 2004) led to the hypothesis that facilitating excitability in the ipsilesional M1 or downregulating excitability in the contralesional M1 could enhance either static excitability or plastic processes leading to improved function. Both strategies have been tried in different laboratories leading to promising proof of principle behavioral and physiological results (for review see Fregni and Pascual-Leone, 2006; Hummel and Cohen, 2006; Talelli and Rothwell, 2006). Studies in healthy subjects pointed to the importance of synchronous application of cortical stimulation to M1 and motor training protocols (Reis et al., 2008a). Possible advantages of combining different stimulating modalities have been suggested as well. For example, a combination of facilitatory stimulation of the ipsilesional and inhibitory stimulation of the contralesional primary motor cortices (Vines et al., 2008) or combination of peripheral nerve stimulation applied to the paretic hand and facilitatory stimulation of the ipsilesional M1 (Celnik et al., 2009) in association with training have shown potential benefits in patients with chronic stroke. Invasive cortical stimulation through epidural electrodes over the primary motor cortex may have similar effects and has been proposed after stroke in humans and in animal models (Brown et al., 2008; Plow et al., 2009). While most reports seem to point to benefits of these techniques, it should be kept in mind that negative results are often underreported and that results from well-controlled, multicenter clinical trials under way are still not available.

Fig. 1.1.

The currently described influences of other brain areas on the output of the primary motor cortex (M1) are shown. Open arrows denote facilitation, while filled arrows denote inhibition. In many cases the influence shown represents a net effect of several specific interactions, whose details are discussed in the relevant section of the text and are shown in subsequent figures. These influences include projections from motor areas in the ipsi- and contralateral hemispheres and the effects of afferent sensory input. PMd = dorsal premotor cortex; PMv = ventral premotor cortex; SMA = supplementary motor area; PPC = posterior parietal cortex; CBL = cerebellum; THAL = thalamus; PNS = peripheral nervous system. (Figure reproduced with permission from Reis et al., 2008b.)

Manipulation of somatosensory input elicits clear effects on somatosensory as well as motor function. In healthy humans, somatosensory stimulation of median, ulnar, and/or radial nerves at the wrist induces clear increases in fMRI activation and cortical excitability in the stimulated hand motor cortical representations (Ridding et al., 2001; Kaelin-Lang et al., 2002; Wu et al., 2005; Conforto et al., 2010) while hand anesthesia induces increases in excitability and improved tactile discrimination in the non-anesthesized hand (Werhahn et al., 2002a, b). In stroke patients, somatosensory stimulation of the paretic limb (Conforto et al., 2002; Sheffler et al., 2006; Celnik et al., 2007) and anesthesia of the non-paretic hand (Floel et al., 2004) show comparable short-lasting behavioral and electrophysiological beneficial effects on paretic hand function, consistent with the documented correction of abnormalities in interhemispheric interactions between the primary motor cortices (Murase et al., 2004; Floel et al., 2008). Results from studies using transcutaneous electrical stimulation are consistent with those carried out using peripheral nerve stimulation and should therefore be considered in neurorehabilitation.

More information is becoming available on the neural mechanisms underlying recovery of motor function and neuroplasticity after stroke (see for example Prabhakaran et al., 2008; Swayne et al., 2008; Marshall et al., 2009). An emerging body of evidence is providing new insight into the interregional interactions between the premotor and parietal areas and primary motor cortex in healthy individuals, as well as strategies to modulate the strength of these interactions (Koch et al., 2006; Koch and Rothwell, 2009). It is only reasonable to expect the development of newer interventional proposals based on the emerging understanding of these mechanisms.

CONCLUSIONS

The last decade saw impressive improvements in our understanding of the ability of the CNS to reorganize in response to changes in the environment and lesions. This understanding resulted in parallel gains in our insight into mechanisms of both action and cognition in health and disease. Understanding of these neuroplastic principles is evolving into the development of more rational, hypothesis-driven strategies to promote recovery of function and will likely result in improvements in patient care along the bench to bedside translational pipeline (Cheeran et al., 2009).