The Mirror Neuron System: A Neural Substrate for Methods in Stroke Rehabilitation

Abstract

Mirror neurons found in the premotor and parietal cortex respond not only during action execution, but also during observation of actions being performed by others. Thus, the motor system may be activated without overt movement. Rehabilitation of motor function after stroke is often challenging due to severity of impairment and poor to absent voluntary movement ability. Methods in stroke rehabilitation based on the mirror neuron system—action observation, motor imagery, and imitation—take advantage of this opportunity to rebuild motor function despite impairments, as an alternative or complement to physical therapy. Here the authors review research into each condition of practice, and discuss the relevance of the mirror neuron system to stroke recovery.

Overview

Stroke is the leading cause of disability among adults,¹ with upper-extremity hemiparesis underlying most functional impairments following stroke.² Recovery may be enhanced by intensive physical therapy aimed at the reorganization of function in damaged neural networks to minimize motor deficits and develop new strategies in motor learning.³ Rehabilitation methods aim to promote adaptive plasticity of structure and function in the undamaged brain toward recovery, driven by motor training.⁴ In patients with poor motor ability, however, participation in physical therapy may be limited, and it may be a challenge to provide relevant input for experience-dependent neural plasticity for neurorehabilitation, repair, and recovery.

One way to overcome these limitations is suggested by the putative mirror neuron system, a parietofrontal neural network active both when we perform an action and when we observe a similar action being performed by others.^5,6 Based on the features of mirror neurons described by single-cell recordings in the monkey’s ventral premotor cortex (area F5) and rostral inferior parietal lobule (IPL),^7–9 converging behavioral, neurophysiological, and brain imaging data suggest a comparable, distributed mirror neuron system in the human brain that is active during both action observation and execution. By representing observed actions in the motor cortex, the mirror system may serve as an alternative means to access the motor system after stroke despite impairments, to rebuild voluntary motor function. A concentrated, interdisciplinary research effort into mirror neurons and the mirror neuron system has improved our understanding of this exciting opportunity. This review highlights efforts to use conditions of practice that engage the mirror system in stroke neurorehabilitation, including action observation, motor imagery, and imitation.

Action Observation

Landmark evidence for a mirror neuron system that is active during both action observation and execution in humans came from a transcranial magnetic stimulation (TMS) study of increased motor cortex excitability during observation of grasping actions.¹⁰ A large volume of subsequent TMS data^11,12 support and extend this initial finding: When we observe actions, our motor system simulates those actions under threshold; increased corticospinal facilitation is specific to muscles used to execute the observed actions¹³ and is temporally correlated with kinematic landmarks of the observed actions.¹⁴ Neural activity associated with action observation using brain imaging methods includes visual processing areas (occipital, temporal, and parietal cortices) and the core mirror neuron system (Figure 1), which comprises motor-related brain regions, including the pars opercularis of the inferior frontal gyrus (IFG), ventral premotor cortex (PMv), and rostral IPL.^15,16 Thus, action observation appears to activate the motor system similarly to execution,¹⁵ generating an internal representation of action that may be a target for stroke neurorehabilitation.

Figure 1.

The core mirror neuron system, including the inferior parietal lobule (IPL), ventral premotor cortex (PMv), and inferior frontal gyrus (IFG).

Recent work demonstrates that action observation drives reorganization of motor representations in the primary motor cortex (M1) to form a motor memory of the observed action.¹⁷ In a recent study, Stefan et al¹⁷ determined a baseline directional bias of thumb movements evoked by TMS over M1 and then showed that observation of repetitive thumb movements oriented opposite this bias changed the direction, net acceleration, and excitability of agonist muscles in favor of the observed actions. This replicates motor training–induced motor memory formation in M1,¹⁸ but note that here, a motor memory resulted from mere action observation. On replicating the experiment in older people (mean age 65 years compared with mean age 34 years), however, motor training or action observation alone did not significantly change TMS-evoked thumb movements from baseline; only combined motor training and action observation led to new motor memory formation.¹⁹ A similar protocol was tested with 8 participants with chronic stroke,²⁰ resulting in motor memory formation by combined motor training and congruent but not incongruent action observation. These results support the use of action observation to augment physical therapy after stroke.

The success of using action observation to drive neural reorganization and repair in stroke recovery, especially in patients with poor motor ability, may depend on the neural integrity of action representations in the motor system, and thus, overlap in the location of activity during action observation and execution may be important. Dinstein et al²¹ used functional magnetic resonance imaging (fMRI) adaptation to assess neural activity selective for observed and/or executed hand actions and found repetition suppression in the anterior frontal sulcus, PMv, anterior IPL, and superior and posterior intraparietal cortices for repeated observation or repeated execution of the same hand action. Although this study did not find adaptation to repeated observation and execution of the same action, recent work by Chong et al²² demonstrated such cross-modal adaptation in the right IPL. Furthermore, Gazzola and Keysers²³ assessed overlap during observation of object-grasping actions, inclusively masked with action execution and exclusively masked with viewing of scrambled images, to reveal relatively right–left overlap in the temporal, parietal, and 2 frontal nodes, including strong activity in the mirror system. An updated analysis to determine direct overlap for action observation and execution²³ used unsmoothed single-participant data from the same study and found shared voxels in the PMv (BA6/44) and inferior parietal cortex, dorsal premotor cortex (PMd), supplementary motor area (SMA), middle cingulate cortex, somatosensory cortex (BA3/2, parietal operculum), superior parietal lobule (SPL), middle temporal cortex, and cerebellum. For the premotor, somatosensory, and parietal activity, shared voxels were more numerous in the left hemisphere. Direct overlap of neural activity in these regions during action observation and execution suggests a substrate that may be accessed after stroke to rebuild motor function by action observation, when motor training is limited by impaired voluntary capability.

Action representations in the mirror system assemble within a distributed neural architecture for goal-oriented behavior that is described by a hierarchical model of the motor system defining action at 4 levels: (1) intention, or the long-term action goal; (2) goal, or the short-term object goal needed to realize intention; (3) kinematics, or the shape of the hand and movement of the arm in space and time; and (4) muscle patterns used to execute the goal.²⁴ For observed actions, the mirror system appears to map intention at PMv based on kinematic and contextual cues^25,26 and goal at IPL based on hand–object interactions.²⁷ These features of action representation and how observed actions are mapped onto the motor system of the observer may be useful to refine conditions of practice based on the mirror system to drive stroke recovery, especially related to lesion location. Underscored is the importance of instructions given to patients with stroke who use these methods—for example, directing attention to the end or means of observed actions may differentially recruit the mirror system and related brain systems.²⁸ The significance of instructions for methods in stroke rehabilitation based on the mirror system will be a recurrent theme in this review.

Observation With Intent to Imitate

Because the key rationale for action observation in stroke rehabilitation is to provide a means to recover lost function within the damaged motor network, the important condition of practice is observation with intent to imitate (OTI) the observed action. OTI may prime the motor system for execution—for example, finger movements are produced faster in response to an imitative cue compared with symbolic or spatial cues.²⁹ A recent meta-analysis³⁰ reviewed OTI and established a lack of current data supporting the use of this condition of practice in stroke rehabilitation, noting a need for studies in nondisabled persons to determine associated neural activity and studies in stroke patients with different lesion sites to determine how this response changes with brain damage. However, several studies included in or subsequent to the review by Pomeroy et al³⁰ address 3 factors important to inform clinical applications of OTI: (1) neural activity during OTI, (2) neural activity common to OTI and action execution, and (3) neural activity different to OTI and passive observation.

Regarding neural activity during OTI, the available description is limited by both study design and data analysis methods. Many studies contain an imitation condition in blocked design fMRI but do not isolate the observation and execution components of the task in design or analysis. From the available data,^31–34 we see that OTI activation includes the precentral gyrus (PMv/PMd); SPL, IPL, and intraparietal sulcus; middle frontal gyrus; and cerebellum.

Although neural activity during OTI is relevant to the application of this condition of practice in stroke rehabilitation, the imperative response may be the overlap in activity during OTI and action execution, similar to action observation. Buccino et al³¹ provide the only available estimate of coincident activity during OTI and execution of guitar chords, and overlap includes pars opercularis of the IFG, rostral IPL, middle frontal gyrus, and cerebellum.

Finally, it is relevant to compare neural activity as modulated by task instruction, or the difference between OTI and observation for another purpose. Two early PET studies^32,33 compared OTI and observation to later recognize meaningful and meaningless hand actions and found that OTI activates the parietal and frontal mirror systems, middle frontal gyrus, and cerebellum. Zentgraf et al³⁵ found that observing gymnastics movements with intent to imitate compared with observing to judge the quality of performance revealed stronger activation of SMA, likely because of the requirement to transform the observed action sequences to body-centered coordinates. Other recent studies^31,34,36 found stronger mirror system activity by OTI relative to observation but did not report the specific anatomy for the contrast.

Overall, neural activity during OTI includes the parietal and frontal mirror systems, middle frontal gyrus, and cerebellum. The middle frontal gyrus has been implicated in top-down attentional processes, working memory, and executive motor control, and in this context may reflect selection of motor acts for imitation of observed actions.³¹ The cerebellum is involved in integration of sensory information, coordination, and motor control based on feedback that adjusts cortical forward and inverse models of action,³⁸ generated here during action observation.

Recent theoretical³⁹ and empirical⁴⁰ evidence supports the use of OTI to augment physical therapy and rebuild motor function after stroke. In the first-ever clinical trial of observation in stroke rehabilitation,⁴⁰ 8 participants with moderate poststroke hemiparesis combined OTI with repetitive motor training of the observed actions for 90 minutes per day. Actions of increasing complexity were observed and imitated each day for 18 days. A control group of 8 participants poststroke combined motor training with observation of geometrical symbols and letters. Significant functional improvement on standard scales occurred for combined action observation and motor training compared with controls despite a stable pretraining baseline and was maintained at 8 weeks posttraining with continued basic physiotherapy. Pretraining and posttraining, participants performed an independent object manipulation task during fMRI. The data revealed a significant increase in activity of the bilateral PMv and superior temporal gyrus, contralesional SMA, and ipsilesional supramarginal gyrus posttraining in the experimental group as compared with the control group. Thus, combined OTI and motor training provides significant improvement of motor function and increased neural activity in motor-related brain regions that also comprise the mirror neuron system.

Motor Imagery

Motor imagery is a dynamic state of internal action representation without overt motor output,⁴¹ and may thus provide access to the motor system after stroke in patients with poor voluntary motor ability. An early meta-analysis of imagery in sports psychology⁴² indicated that, broadly defined, mental practice of motor skills produced a positive effect (general effect size 0.48) compared with no practice, and urged that further research on imagery in sports examine the variables mediating relations between mental practice and motor performance. Motor imagery obeys principles of motor control (for a review, see de Vries and Mulder⁴³) and associated autonomic response,⁴⁴ relies on the same neural substrate as action execution,⁴¹ and drives neural reorganization similarly to motor training.⁴⁵ Because the purpose of motor imagery after stroke is to drive recovery of voluntary motor function, the important condition of practice is kinesthetic motor imagery, in which actions are imagined from a first-person perspective, in terms of how it feels to move, rather than visual imagery of action from the third-person perspective. Neural activity associated with kinesthetic motor imagery includes the parietal cortex (SPL, IPL, supramarginal gyrus), somatosensory cortex, frontal motor areas (PMv, PMd, precentral gyrus, IFG, SMA), dorsolateral prefrontal cortex, superior temporal gyrus, anterior cingulate cortex, basal ganglia, and cerebellum,^16,39 thus providing distributed access to the motor system that is partially congruent with action observation and execution, and includes the mirror neuron system.

Within this network, the activation of M1 may be especially relevant to rebuild motor output and recover lost function after stroke, but recent reviews^46,47 find inconsistent M1 activation by motor imagery and debate if M1 involvement is important to stroke rehabilitation because motor imagery may rebuild planning and preparation areas rather than execution. Recent work suggests that M1 may initially become active during motor imagery but is quickly suppressed by SMA.⁴⁸ However, motor imagery results in corticospinal facilitation of the specific muscles used to execute the imagined action⁴⁶ that is preserved in older people⁴⁹ and after stroke.⁵⁰ Furthermore, motor imagery after stroke increased hand motor map area and volume evoked by TMS over M1, an effect that was greater for the ipsilesional hemisphere and thus partly balanced abnormal asymmetry of excitability between hemispheres at rest.⁵⁰ Similarly, motor imagery activated M1 in a study of patients with subcortical stroke, and laterality of this activation correlated with motor scores, in that patients with more motor recovery showed greater ipsilesional M1 activity.⁵¹

Use of motor imagery as a condition of practice in stroke rehabilitation may be complicated by an inability to perform motor imagery because of lesion location; patient failure of compliance or concealed use of an alternative strategy of mental practice; lack of accuracy or temporal coupling of imagined movements, termed chaotic motor imagery; or the failure to suppress overt action.⁴⁷ Thus, inclusion criteria, assessment of motor imagery ability, and monitoring of cognitive strategy are important to clinical applications of motor imagery after stroke. Lesions in the right posterior parietal cortex⁵² or left frontal cortex⁵³ are shown to disrupt motor imagery, whereas patients with other varied lesion sites retain the ability to imagine actions⁵³ and may even show a hemiplegic advantage of more accurate motor imagery with the affected limb in chronic stroke,⁵⁴ possibly because of ongoing motor imagery of impossible movements. Motor imagery ability may be assessed by using a questionnaire—for example, the Movement Imagery Questionnaire–Revised second version⁵⁵ and Kinesthetic and Visual Imagery Questionnaire⁵⁶ assess quality and vividness of imagined actions, and the Chaotic Motor Imagery Assessment⁵⁷ tests mental rotation, timing, and Fitt’s law to control for cognitive strategy.

Given these factors, motor imagery has been used effectively to augment physical therapy in acute,⁵⁸ subacute,⁵⁹ and chronic stroke^60,61 randomized controlled and controlled clinical trials.^62,63 Page et al^59,61,64,65 combined motor training with 30 minutes of audio-guided motor imagery 2 days per week for 6 weeks and found reduced motor impairment and increased use of the affected limb compared with motor training and audio-guided relaxation. Liu et al⁵⁸ compared motor imagery to demonstration-based training of daily activities (eg, folding laundry, taking transportation) by action sequencing, problem solving, and evaluative practice for 1 hour, 5 days per week for 3 weeks, and found better relearning, retention, and transfer to untrained tasks by motor imagery. Butler and Page⁶⁶ further measured the efficacy of combined motor imagery and constraint-induced movement therapy to motor imagery or constraint-induced movement therapy alone in 4 participants poststroke and found promising improvement in motor function and associated cortical reorganization in the combined condition. Finally, Page et al⁶⁷ recently combined repetitive, task-specific practice with motor imagery in 10 participants poststroke and showed postintervention reorganization of cortical activation maps for the affected hand by fMRI. Thus, structural and functional changes of the motor cortex associated with motor training are also found with motor imagery after stroke, similarly to action observation, including synaptic and cortical map plasticity. These studies provide a solid theoretical basis and beginning empirical support for the use of motor imagery as a complement to physical therapy in stroke rehabilitation (for a review, see Mulder⁶⁸).

Imitation

The basic circuit for imitation matches that for action observation, converting a visual representation of observed actions from the superior temporal sulcus to an object-oriented kinesthetic representation at the parietal mirror neuron system and onto a motor program related to action goal at the frontal mirror system,⁶⁹ which outputs motor commands for imitation. At the same time, an efference copy of the motor command is fed to the parietal cortex as a forward model of predicted sensory feedback that is compared within the circuit to actual sensory feedback for online motor control.⁷⁰ The HMOSAIC model of motor control⁷¹ provides a theoretical framework consistent with this flow of information for action execution,⁷² observation, and imitation,⁷³ with paired inverse-forward models in parallel and nested within a hierarchical structure that predicts sensory feedback at different levels of the action hierarchy described above.²⁴ In this framework, the mirror neuron systems may support imitation within a larger set of brain regions responsible for working memory and motor preparation, including the dorsolateral prefrontal cortex, PMd, pre-SMA, and SPL.^31,69,74

Within the basic imitation circuit of the superior temporal sulcus and parietofrontal mirror system, the IFG seems particularly important. Activity in IFG increases in a stepwise manner from action observation to execution to imitation,⁷⁵ and activity in IFG and PMd is greater for imitation of goal-directed actions compared with actions with no explicit goal.⁷⁶ A causal relationship between IFG and imitation was established by Heiser et al,⁷⁷ who showed that repetitive TMS to IFG disrupts imitation of finger movements but not a control visuomotor task.⁷⁷ A recent review of IFG and imitation⁷⁸ found functional segregation of the pars opercularis of IFG into a dorsal sector active during action observation and imitation, and a ventral sector active only during imitation, and suggests that it is this ventral sector that inputs an efference copy of the motor command to the forward model.

A recent review of the mirror neuron system for neurorehabilitation³⁹ describes the features of imitation that promote reorganization and recovery of motor function after stroke, including the following: (1) distributed neural activity in response to multiple sensory inputs allows many options for proper activation of the network; (2) activation of the mirror neuron system results in corticospinal facilitation even without overt motor output; (3) the network is associated with learned, ecologically valid actions rather than fragmented movements typical to rehabilitation; and (4) empirical data suggest recovery of function by motor imagery, to which imitation adds the components of action observation and execution to reinforce rehabilitation.

Further theoretical support for imitation as a condition of practice after stroke comes from a 2-route model for imitation of meaningful and meaningless actions⁷⁹ that includes (1) a direct route to transform novel actions into motor output and (2) an indirect, semantic route for imitation of known actions. On visual analysis, meaningless actions that do not have stored goals bypass semantic memory to directly activate short-term working memory for imitation (route 1), associated with activity in the right parieto-occipital junction. Meaningful actions with goal representations activate long-term semantic memory before accessing working memory for imitation (route 2), associated with activity in the left inferior temporal gyrus, or may instead access the direct route. Direct and indirect routes for imitation may be supported by dorsal and ventral streams,⁸⁰ respectively, thus providing further distribution of access to the motor system for stroke rehabilitation.

A related distinction is made between imitation, goal emulation, and movement priming.⁸¹ Goal emulation is the attempt to achieve an observed goal by any means, and movement priming is the copying of body movements but not as learned means to a goal. Combined, these forms of social learning provide the ends-and-means structure of imitation that may be distinguished in the frontal and parietal mirror neuron system.^25,27

To recover motor function after stroke, methods in rehabilitation based on the mirror neuron system may be considered as driven toward eventual imitation as a combined action observation, motor imagery and execution condition. It is interesting to note that virtual reality interventions used to promote recovery from stroke (for a review, see Holden⁸²) may work in part by engaging the mirror neuron system as a form of imitation, rewarding partial movements by reinforcing the motor circuitry responsible for actual execution of observed actions.

Considerations

It is important to consider the role of mirror neurons as part of much larger systems in the brain and, likewise, how conditions of practice based on the mirror systems fit into a framework of action-oriented training to promote recovery from stroke. It follows that stroke deficits are not only a manifestation of lesion location, but also represent the ability of the rest of the brain to maintain function. Fregni and Pascual-Leone⁸³ offer a conceptual framework for stroke recovery research that may be useful in the study of action observation, motor imagery, and imitation in rehabilitation. Key to this framework is that functional recovery after stroke is essentially motor learning within a partially disrupted network. Motor learning is operative during recovery and interacts with rehabilitation,⁸⁴ and principles of motor learning may be applied to conditions of practice to drive relearning of motor skills after stroke. Motor learning can occur without action execution; for example, Mattar and Gribble⁸⁵ show that participants who observed an actor learning to move a robotic arm to targets against a clockwise force field executed the same task more accurately than those who did not observe learning, observed learning of a counterclockwise force field, or observed attempted learning of a randomly varying force field⁸⁵; thus, movement strategy was acquired based on specific action observation. Unrelated arm movement during observation impaired participant performance, whereas basic mathematics did not, suggesting that the motor system but not attention must be unoccupied for motor learning by observing. A related study by Cross et al⁸⁶ showed that action representations in the mirror neuron system can be acquired by motor or observation learning.

Of further importance to the framework for stroke recovery research is that corticospinal output must be adequate to allow recovery of motor function; recovery of function involves shifts of distributed contributions across the motor system; and the time course of recovery involves changes in intrahemispheric and especially interhemispheric interactions to first minimize damage and later promote functional recovery. Thus, we must consider lesion location and residual motor function; functional specialization, integration, and plasticity or reorganization of motor function; and optimal timing and duration of rehabilitation, when applying conditions of practice based on the mirror systems to stroke recovery. Although there is some beginning work in these important areas,^87–89 considerably more research is needed to develop evidence-based guidelines for neurorehabilitation that best harness the opportunity for improved outcomes after stroke.

Brain imaging studies can help determine which patients will benefit most from one method of stroke rehabilitation or another, and they already highlight the importance of incorporating information about lesion location and volume into the planning of stroke treatment trials.⁹⁰ Lesion data will be useful to develop predictive models of stroke recovery (eg, see Miyai et al⁹¹) to more effectively match patients with specific rehabilitation methods. It follows that any behavioral result in stroke research should be interpreted with consideration of lesion information as derived from brain imaging. Given the mirror neuron system pathway—from the visual cortex to the parietal and premotor mirror system—we may predict that a subcortical lesion of the internal capsule without motor cortex involvement would leave an intact mirror system, useful to stroke rehabilitation. A patient with this lesion profile and at least 25% corticospinal tract integrity may benefit more from the methods discussed in this review than another patient with cortical damage directly affecting the mirror system. Future research that is designed to test such predictions is sorely needed to accelerate the translation of our understanding of the mirror system to clinical practice.

Finally, it is interesting to note that imagery and modeling methods are widely used for motor learning and improved performance in practical fields such as sports psychology and motor control, and similar issues arise in such applications regarding specific task and instructions, and individual participant characteristics.⁹² A sound neuroscientific framework for methods that engage the mirror neuron system is similarly needed to advance translational research in sports and motor control. Research of action observation, motor imagery, and imitation in these fields will contribute to the development of methods in stroke rehabilitation; for example, the mirror neuron system appears to respond most strongly during observation of acquired motor skills, such as when ballet dancers observe ballet compared with capoeira⁹³ or when experts compared with novice high jumpers imagine high jumps.⁹⁴ Thus, methods based on the mirror system in stroke rehabilitation may benefit by incorporating actions that are already highly represented in the motor system based on motor experience prior to brain damage and that the patient finds meaningful.

Conclusions

Action observation, motor imagery, and imitation are represented in the same basic motor circuit as action execution—the mirror neuron system—and thus provide an additional or alternative source of information to motor training that may be useful to promote recovery from stroke. Current work on the basic science of mirror neurons and systems will further inform clinical applications in stroke rehabilitation, and continued research efforts to apply these conditions of practice in the clinic will provide further evidence supporting the use of methods based on mirror neurons for stroke recovery.