Human beings are social creatures to the extent that interactions with members of their own species, and especially the ability to understand and infer the intentions and beliefs of others, has become of predominant importance in their daily life. Whether for cooperation or non-cooperation, a core assumption of this viewpoint is that such social interactions spring from a distinction between self and others. It can be argued that at least two hierarchically-organized, overlapping and interacting neural systems have evolved and developed to manage self-other distinctions and hence social interactions [1
]. One system, part of the classic motor system, is more specialized for the preparation and execution of motor actions that are self realized and voluntary, while the other appears to be more involved in capturing and understanding, at a basic and involuntary level, the actions of non-self or others. For our purposes, actions are defined as sequences of movements that together solve a motor problem [2
] and that involve at least four levels of behavioral complexity: intention, kinematics, goal-object identity, and the physical consequences of the action [1
]. Motor preparation and execution circuitry includes, among others, the premotor cortex, supplementary motor area, sensorimotor cortices, and parts of the inferior parietal cortex. The second system, of which the mirror neuron system (MNS) is part, has been described as the canonical action 'resonance' system in the brain – one that has evolved to utilize or share many of the same circuits involved in motor control [3
]. Mirroring or 'shared circuit' systems are assumed to be important for resonating, imitating, and/or simulating the actions of others. Although no consensus exists, a number of researchers have proposed that shared representations of motor actions, or the action understanding properties of this system, may form a foundational cornerstone for higher order social processes, including motor learning, action understanding, imitation, perspective taking, understanding facial emotions, and empathy [4
]. This means that adopting someone else's viewpoint or perspective at the very least requires that the other's actions be understood
; else no accurate prediction of their behavior can be made.
However, a mirroring system that evolves and is adapted from the classic motor system presents at least three major problems: a development, a correspondence, and a control problem. In terms of the development problem the question is whether humans acquire the ability to mirror by mapping observed onto executed actions? That is, how exactly does a mirroring system arise? Are mirror neurons innate and therefore genetically programmed? Is learning necessary? And, what role does sensorimotor cortex play? A number of studies have indicated that imitation of facial and hand gestures in both human and non-human primates suggest the existence of mirroring systems in infancy [9
]. Likewise, electroencephalography (EEG) and near infrared spectroscopy studies in humans show sensitivity to executed versus observed actions, as well as between live and televised actions [12
] suggesting the existence of mirroring as early as 6–7 months of life. However, none of these studies directly answers the development questions posed. On the other hand, computational models of mirroring activity propose that sensorimotor transformations, via Hebbian learning, can in fact mediate such development.
In terms of the correspondence problem the question is how does the observer agent determine what the observed agent's activation pattern is in order to match it? Or, as Brass and Heyes [17
] stated the problem with respect to imitation, "When we observe another person moving we do not see the muscle activation underlying their movement but rather the external consequences of that activation. So, how does the observer's motor system 'know' which muscle activations will lead to the observed movement?" Resonance becomes particularly difficult when the observer and observed do not share the same embodiment and affordances, that is, they do not share all "action possibilities" latent in the environment. One partial solution to this problem, of course, exists in the implicit nature of a mirroring system, i.e., a system that evokes motor representations by movement observation
. That is, if motor actions already exist as part of the observer agent's movement repertoire then observation of action, even when partially triggered, can be sufficient to evoke the representation. This solution clearly makes sensorimotor transformations, as part of a mirroring system, necessary for solving such a correspondence problem.
Finally, in terms of the control problem, the issue arises because an efficient mirroring system ought to be turned on only when needed. However, it has been shown repeatedly that activation of internal motor representations via observation occurs automatically. Neuroimaging studies, for example, show that simple passive observation is enough to generate motor activation. The question then is how to control a system for efficiency when it is turned on automatically? Or, as others have stated the problem succinctly: "Why don't we imitate all the time?" The existence of neural inhibitory and monitoring mechanisms as partial solutions to this control problem has been acknowledged [3
], although the specific anatomical implementation of such mechanisms is unknown. Brass and colleagues [18
], for example, found that the fronto-median cortex and the right temporo-parietal junction were activated when an instructed movement had to be executed during observation of an incongruent movement. The implication being that high level areas are involved in inhibition of imitative response tendencies. Another solution centers on phasic changes in oscillatory EEG activity as inhibitory control mechanisms. This is consistent with the role of sensorimotor cortex as a critical region for mirroring based on its common output path role in motor and simulation-based representations. More specifically, we hypothesize that oscillatory activity, such as mu rhythms in sensorimotor cortex, play a key role in controlling mirroring processes.
Mirroring activity can be conceptualized as occurring in a gradient. At one end of the spectrum, the mimicry of another individual's postures, facial expressions, vocalizations, movements and mannerisms is often executed in the absence of awareness, as occurs in the chameleon effect, motor empathy, motor contagion, or emotional contagion [19
]. At the other end of the spectrum, it has been suggested that simulation based on mapping of observed actions onto one's own motor system necessitates the interaction with semantic/cognitive circuits for conscious action understanding to occur [20
]. We conceptualize this spectrum of action understanding as reflecting four levels of behavioral complexity, i.e., intentions, goals, patterns of muscle activation, and kinematics, as has been suggested by Hamilton and Grafton [1
]. Furthermore, we argue that these levels of processing can be mapped onto differences in activation in different components within a 'core' and an 'extended' mirror neuron system (see Figure ). Although it remains to be definitively shown, differential activation of the various components of this mirroring system most likely result as a function of the task, working memory, motivational and/or attentional factors involved. In this paper, we argue from an anatomical, physiological, modeling, and functional perspectives that one critical component of an 'extended' mirror neuron system is sensorimotor cortex. This region is necessary not only for computing the patterns of muscle activation and kinematics during action observation but provides potential answers to the development, correspondence and control problems in mirroring.
Figure 1 Schematic of areas in the human brain that contain mirror neurons (inferior parietal lobule and inferior frontal gyrus) and make up the 'core'system. The 'extended' mirror neuron system involves additional brain areas, e.g., insula, middle temporal gyrus, (more ...)
The 'core' MNS
The mirror neuron system has been widely defined as consisting of three interrelated areas: ventral premotor area (PMv) of the inferior frontal gyrus (area F5 in monkeys), parietal frontal (PF) in the rostral cortical convexity of the inferior parietal lobule (IPL), and the superior temporal sulcus (STS) (see Figures , and , as well as Table for a description of these areas). The mirror neuron circuit in monkeys [4
] begins in the rostral part of the superior temporal sulcus, although no mirror neurons per se
have been reported in this area. Information is then thought to flow to the parietal frontal area on the rostral cortical convexity of the inferior parietal lobule. A subset of the cells in this region has mirror properties: i.e., they discharge both when the monkey executes as well as observes an action. Parietal frontal area, in turn, sends projections to area F5 of the ventral premotor area, where a subset of cells (10–20%) exhibits mirror properties. Thus, the core mirror neuron system would be defined as those areas that contain mirror-like neurons, which at this point includes primarily the rostral convexity of the inferior parietal lobule or parietal frontal area and ventral premotor area.
Anatomical view of a human brain showing areas involved with the mirror neuron system.
Anatomical view of a macaque monkey brain showing areas involved with the mirror neuron system.
Abbreviations and functional descriptions of anatomical areas
Single unit studies in the premotor cortex of macaque monkeys indicate that neurons in area F5, particularly in the caudal portion of the inferior frontal gyrus (IFG), are indistinguishable from neighboring neurons in terms of their motor properties and discharge in response to executed and observed actions [23
] (for a review see [4
]). The implication is that when a monkey observes an action, particularly one that is in its motor repertoire, a subset of neurons in this region 'mirrors' the activity and represents the motor action in its own premotor cortex, revealing a type of observation/execution matching system. This type of observation/execution activity has been shown to be selective for goal-directed, meaningful actions supporting the idea that actions are organized with respect to distal goals [24
]. More recently, another subpopulation of neurons in the same area of the monkey has been found that discharges both when the animal performs a specific action as well as when it sees or hears the same action performed by another individual [25
]. That is, these cells represent in an individual's motor cortex not only the execution of an action (motor representation) but also the 'observation' of that action performed by others (visual representation), as well as its auditory correlates (auditory representation). In other words, auditory mirror neurons allow for a mapping of specific heard actions onto the motor programs for executing the same actions.
Individual human mirror neurons cannot be studied directly except under unusual circumstances [27
]. Nonetheless, the evidence suggests that the motor related part of Broca's region is located in the caudal portion of the inferior frontal cortex, in what is Brodmann's area 44, and there appears to be a homology between area F5 in the monkey and area 44 in humans. Area 44 is involved in interfacing external information about biological motion and internal motor representation of hand/arm and mouth actions [28
]. Hence, the existence of an analogous mirroring system in the homologous human brain regions has been supported by indirect population-level measures such as electroencephalography [12
], magnetoencephalography [35
], transcranial magnetic stimulation [36
], positron emission tomography [37
] and functional magnetic resonance imaging [19
]. Fadiga and colleagues [36
], for example, found that motor evoked potentials over motor cortex were enhanced in response to transcranial magnetic stimulation when subjects observed another individual performing an action relative to when they detected the dimming of a light. Iacoboni and colleagues [39
] measured blood flow in Brodmann's area 44 and found increases during the observation and performance of actions. Other studies have reported activations with similar properties in the parietal cortex [40
], as well as the superior temporal sulcus [42
]. In general, the human mirror neuron system appears active during the performance and observation of the same action and is hypothesized to be necessary for imitative learning [44
], comprehending the actions of others [24
], understanding the goal of another's actions [46
], interpreting facial expressions [19
], and exhibiting empathy [19
The 'extended' MNS
It has been shown that we activate our own motor, somatosensory, and nociceptive representations while perceiving the actions of others, while at the same time activating representations of our own emotional states as well as facial expressions while witnessing others' emotions [48
]. At minimum, this activation of shared representations for action and emotion requires a variety of anatomical and functional circuits that together might be called the 'extended' mirror neuron system. Undoubtedly, the core mirror neuron areas, as described previously (see Figure ), are anatomically connected with many other regions that contribute significantly to the subsequent elaboration of the information [49
]. Those regions may themselves not contain mirror neurons per se
, such as the superior temporal sulcus, but the level of transformation performed on the data would make them critical to the outcome and part of an extended mirroring process.
The arguments as to why the superior temporal sulcus, despite the lack of mirror neurons, is considered part of a mirror neuron system are both anatomical and functional [51
]. It is an area that contains neurons that respond to biologically relevant actions of the head, body, and eyes, as well as to static pictures that merely imply biological motion [52
]. Furthermore, this area is reciprocally connected to the parietal frontal area in the inferior parietal lobule. However, the functional significance of the mirror neuron system has to be understood in its connections to many other neural systems [20
]. Thus, the degree to which brain areas in these other systems play a critical role in action understanding or in any of the processes attributed to the core mirror neuron system would define their inclusion as part of an extended circuit.
The extant evidence supports inclusion of a number of areas into an extended definition of the mirror neuron system. For example, the subjective sense of how one feels is theorized to be based upon anterior insula representations of the body. This is assumed to provide a foundation for emotions and perhaps even for self-awareness that could allow for simulation of future actions, in order to use the feelings generated by the simulation to guide decision making [53
]. Singer and colleagues [54
] found, in a functional magnetic resonance imaging study, that empathy for pain involves simulating the unpleasant, aversive qualities of the pain (the motivational significance of pain) but not its precise somatic characteristics. In another study, Saarela and Hari [55
] used photos of facial expressions from chronic pain sufferers which varied in the intensity of depicted suffering. Not only were bilateral anterior insula, left anterior cingulate, and left inferior parietal lobe activated, but the amount of these activations correlated with subjects' estimates of the intensity of observed pain. Clearly, the insula has an important role in mirroring and should be considered part of the extended mirroring system. Likewise, observation-evoked motor activity, as well as mirror-type activity, has been reported in dorsal premotor cortices [56
], while the middle temporal gyrus (MTG) and adjacent superior temporal sulcus are often found to show augmented blood-oxygen level dependent (BOLD) responses during action execution and action observation [58
]. Finally, and most relevant to the argument in this paper, primary and secondary motor and somatosensory cortices often contain voxels active during both action execution and observation/listening [13