Nonverbal communication is a crucial component of human social behavior, but its neural mechanisms are poorly understood. The ability to understand others’ mental states from their facial and bodily gestures allows us to respond effectively during social communication. Gallese and Goldman (1998) proposed a simulation theory of action understanding to account for the complexity of this process. Under this model, on observing an action, the observer subconsciously and automatically employs a specialized neural circuitry to simulate the action using their own motor system, in turn activating mental states associated with execution of the action, and providing insight into the mental state of the actor. The neural substrate of the simulation theory is proposed to be the mirror neuron (Gallese and Goldman, 1998).

Mirror neurons were first discovered in the motor areas of the monkey brain (di Pellegrino et al., 1992). They were observed to fire during both execution and observation of actions, such as grasping an object, putting it in mouth or breaking it. Moreover, the sensory modality by which the action was experienced did not seem to matter for a subset of these neurons: they were triggered by the sound of the action, even when the action was not seen (Kohler et al., 2002). The implication of these findings was that the mirror neurons could be coding the representations of the actions, allowing for recognizing the movements involved in an action and inferring the intention behind the action. Evidence for a similar mirroring mechanism in the human brain has come from functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) studies (Caspers et al., 2010; Molenberghs et al., 2012) as well as single neuron recordings during surgery in humans (Mukamel et al., 2010).

In addition to metabolic brain imaging and in vivo cellular studies, Electroencephalography (EEG) studies have measured mu rhythm desynchronization to infer mirroring activity. Mu rhythm, characterized by 8–13 Hz oscillations detected over the sensorimotor area, is mostly associated with the functions of the sensorimotor cortex (Niedermeyer, 2005). Increased mu rhythm power indicates physical inactivity and resting, with movement execution as well as observation leading to its suppression (Hari and Salmelin, 1997; Cochin et al., 1998, 1999; Fecteau et al., 2004; Muthukumaraswamy et al., 2004; Lepage and Théoret, 2006). Due to the responsivity of the mu rhythm to action observation, it has been proposed to reflect mirror neuron activity related to viewing of biological action with or without object interaction, including finger movements (Babiloni et al., 1999; Cochin et al., 1999) and hand grip movements (Muthukumaraswamy et al., 2004), as well as hearing sounds that are linked to actions, such as piano melodies (Wu et al., 2016).

Since the initial discovery of mirror neurons, research has focused on their potential role in social cognitive processes that rely on an ability to understand actions and intentions, such as empathy. Similar activity in the brain regions observed in fMRI during execution and observation of facial expressions has been suggested to provide evidence for the existence of a single mechanism of action representation which allows people to empathize with others (Carr et al., 2003). In order to further the knowledge about the role of the mirror neuron network (MNN) in social emotional information processing, a group of researchers used EEG mu suppression as a proxy of the MNN to investigate the network’s sensitivity to emotional information using body parts in painful and non-painful situations, and found greater mu suppression in the painful compared to non-painful conditions (Yang et al., 2009; Cheng et al., 2014; Hoenen et al., 2015). In contrast to findings which suggest a heightened sensitivity of the mu rhythm to emotional information, others found similar levels of mu modulation during gender discrimination and emotion recognition tasks which entailed viewing point-light displays of human figures’ walk (Perry et al., 2010b). Facial expressions have been used as stimuli in EEG mu suppression research only in a handful of studies (Moore et al., 2012; Cooper et al., 2013; Rayson et al., 2016, 2017; Moore and Franz, 2017). Further research using different types of facial movements depicting varying levels of emotional information as the visual stimuli is necessary to investigate the differential sensitivity of the sensorimotor cortex to emotion-related information processing.

It is crucial to note that findings from some mu suppression studies indicate that mu can easily be confounded with occipital alpha activity, yielding alpha suppression at the central electrodes that is not only similar while viewing biological and non-biological motion, but also more pronounced to biological than non-biological motion when the observed action depicts pain. As pointed out by Milston et al. (2013), most of the studies that have explored the relation between mu suppression and empathy have used stimuli eliciting pain only (e.g., Yang et al., 2009; Perry et al., 2010a; Hoenen et al., 2015). Researchers have highlighted that processes other than empathy, such as attention, may be at work while viewing painful stimuli due to their threatening nature (Hoenen et al., 2013) or salience (Perry et al., 2010a). It may still be difficult to disentangle mu from alpha in tasks using non-emotional biological motion. For example, Aleksandrov and Tugin (2012) did not find any systematic differences in mu suppression to the observation of hand movements, non-biological objects or mental counting. Similarly, Perry and Bentin (2010) observed that alpha suppression at the mu and the occipital areas were very similar to the observation of hand movements toward an object. A recent study conducted by Hobson and Bishop (2016) showed that different types of baseline used to measure mu suppression engage the attention system differently, thus directly impacting the degree of suppression recorded. They found that mu and occipital alpha modulation while viewing hand movements and kaleidoscope movements were consistent with the MNN activity only when the static video of the image that immediately preceded the dynamic video of the image was used as the baseline. Due to the posterior alpha confound associated with attentional processes, the baseline and the control conditions need to be chosen carefully.

The current study aims to contribute to our understanding of the simulation account by investigating the responsiveness of the sensorimotor cortex to emotional and non-emotional facial expressions. Our goal is to examine the differential sensitivity of the mu rhythm to different types of facial movements. To our knowledge, this is the first study to examine the sensitivity of mu rhythm, while controling for occipital alpha activity, to dynamic neutral and emotional facial expressions not depicting pain. A within-trial baseline method was adopted as per Hobson and Bishop (2016): the 1,100 ms static image epoch was used as the baseline for quantifying activity in the subsequent 2,050 ms dynamic image epoch. It was hypothesized that mu suppression would be greater in the: (1) happy, sad and neutral face conditions than the non-biological stimulus condition; and (2) happy and sad face conditions than the neutral face condition, without a corresponding difference in occipital alpha suppression.

The current study investigated the modulation of mu rhythm while participants observed videos of emotional and neutral face movements and non-biological stimulus movements. Mu suppression, but not occipital alpha suppression, was predicted to be greater in the face conditions than the non-biological stimulus condition, with greater suppression in the emotional faces than the neutral face condition. In contrast to our prediction, only the neutral faces were associated with stronger mu activity than that for the non-biological stimulus condition. A lack of difference in mu/alpha band power between the emotional faces and the non-biological stimulus at the central region made it difficult to distinguish mu from posterior alpha modulation during emotional face observation. Greater suppression in the neutral face than the non-biological stimulus condition at the central region accompanied with an opposite pattern at the occipital region suggests that mu rhythm modulation associated with neutral face processing is distinct from the attenuation of the overall alpha activity power associated with information processing and attention. Similar opposing trends of alpha and mu suppression between biological and non-biological movement was observed by Hobson and Bishop (2016). Greater occipital alpha suppression in the non-biological stimulus than the neutral face condition may be explained by low-level visual differences between the two conditions, such as the contrast and the frequency domain information in the stimuli, and/or disparate demands on attention.

In addition to the results from the scalp-recorded EEG activity, source analysis data provide further support for a more localized than an overall difference in the mu/alpha band power between neutral face and non-biological stimulus conditions, suggesting different levels of activity between conditions in the face-related (i.e., fusiform gyrus) and MNN areas, specifically, the primary somatosensory cortex, prefrontal cortex and supplementary motor area. Greater activity in the fusiform gyrus in response to faces than non-biological stimulus was expected as this area responds more to faces than objects (Haxby et al., 2000). The premotor areas, including the supplementary motor area, and the primary somatosensory cortex are the key regions implicated in sensorimotor simulation during action observation (for a review, see Wood et al., 2016) and motor imagery (Burianová et al., 2013; Filgueiras et al., 2018). The premotor cortex has been a primary region investigated in studies of action observation Buccino et al., 2001; Johnson-Frey et al., 2003; Raos et al., 2004, 2007). While the motor representations of actions are stored in the premotor areas, the somatosensory areas may be involved in storing tactile and proprioceptive representations of these actions (Gazzola and Keysers, 2009). In addition to the role of the somatosensory activity in hand actions (Avikainen et al., 2002; Raos et al., 2004), there is evidence for the involvement of somatosensory representations in our ability to simulate basic emotions while observing facial expressions (Adolphs et al., 2000). Wood et al. (2016) review highlights the role of sensory simulation in addition to motor simulation in emotion recognition, pointing to a large overlap between brain areas involved in production and observation of facial expressions. Signaling from the somatosensory cortex to the premotor cortex may be a necessary step for action understanding and imitation (Gazzola and Keysers, 2009). This signaling may explain the significantly less mu/alpha band power present source estimation results show in these two brain areas in the neutral face movement compared to the non-biological stimulus condition.

We offer a number of possible explanations for the EEG results showing the strongest mu suppression to the neutral face movement in the form of mouth opening. Firstly, the results may be attributed to the sensitivity of the sensorimotor cortex to human-object interaction. Most research that has investigated the role of MNN in action observation involves hand and finger movements that almost always suggest some sort of interaction with an object, such as pincer movement with the thumb and the index finger (e.g., Cochin et al., 1999), manipulating objects (e.g., Gazzola and Keysers, 2009), or bringing food to mouth (Ferrari et al., 2003). In addition to limb movements, viewing oro-facial movements has also been observed to induce mu power decrease, with greatest suppression to viewing object-directed actions compared to undirected sucking and biting movements, and least suppression to the viewing of speech-like mouth movements (Muthukumaraswamy et al., 2006). In the present study, the sensorimotor cortex could be engaged by the mouth opening gesture which may have been perceived as an action associated with eating, an action implying interaction with an object (i.e., food), thereby supporting intention understanding (i.e., eating). Second, the MNN may be involved in the recognition of deliberate, voluntary gestures rather than involuntary communicative actions. Yet, mu suppression is reported to be modulated by contextual information, such as the actor’s familiarity (Oberman et al., 2008) or their reward value (Gros et al., 2015), or gaming context in which the hand gestures are viewed (Perry et al., 2011). In addition, viewing facial gestures that do not suggest object interaction or deliberate action also seems to modulate mu rhythm (Moore et al., 2012; Rayson et al., 2016, 2017; Moore and Franz, 2017). Thus, explanations which restrict mu suppression to voluntary or object-related actions are unlikely.

A third explanation is that different types of facial movements may tap different MNN areas. An fMRI experiment conducted by van der Gaag et al. (2007) found bilateral inferior frontal operculum activation to viewing emotional facial expressions but somatosensory activation to neutral movements (i.e., blowing up the cheeks). The authors attributed their findings to distinct processing pathways, more visceral in the former and more proprioceptive in the latter. A similar differential pathway may explain the current findings. Alternatively, if a single mirroring pathway underlies all types of facial movements, greater ambiguity of the action and/or the emotion in the mouth opening image may require the MNN more than the full-blown, easy to recognize emotional expressions. In other words, when the emotion information is presented in high intensity, the cognitive task of recognition may not be demanding enough to activate the MNN, thereby bypassing the whole system, as indexed by the lack of or reduced mu suppression.

Based on recent findings from connectivity research (see, for example, Gardner et al., 2015), our last explanation argues that rather than a global increase/decrease of activity in the totality of the network, a differential modulation of the signaling between the key MNN nodes is more likely to be at work during action observation. There is evidence for the existence of a subgroup of neurons in the human supplementary motor area that is excited by execution, but inhibited by observation of hand grasping actions and facial emotional expressions (Mukamel et al., 2010). These observation-inhibited neurons may be the mechanism for self-other discrimination process related to observing others’ actions, and the strength of their activity may modulate the amount of input from premotor areas to the sensorimotor cortex during action observation (Mukamel et al., 2010; Woodruff et al., 2011). Readily recognizable emotion-related information may activate the observation-inhibited mirror neurons in the premotor areas, leading to less excitatory input to the sensorimotor cortex. On the other hand, neutral facial movements that lack social and emotional information, as in mouth opening, may not activate observation-inhibited neurons as much as easily recognizable expressions do, resulting in stronger excitatory input to the sensorimotor cortex. Thus, in the face of subtle expressions, increased sensorimotor activity may aid action and emotion recognition.

Signaling between and within the key MNN areas during action observation and execution has recently been approached from a Bayesian perspective that suggests the existence of an updating mechanism which continuously attempts to minimize the difference (i.e., the error) between the predicted action and the observed or executed action to achieve an understanding of the most likely cause of an action (Keysers and Perrett, 2004; Kilner et al., 2007b). According to a predictive coding model of mirror neurons, when the mismatch between the predicted and observed actions of others is large due to the unfamiliarity, unusualness and unexpectedness of the observed action, the network generates a new prediction model, resulting in stronger motor activation (Kilner et al., 2007a,b). In line with this account, several studies have reported greater mu suppression in infants during observation of extraordinary actions (e.g., turning on a lamp with one’s forehead or lifting a cup to the ear) compared to ordinary actions (e.g., turning on a lamp with one’s hand or lifting a cup to the mouth), suggesting that as the deviation of the observed action from the expected action increases, motor activation increases (Stapel et al., 2010; Langeloh et al., 2018). In the current study, the unfamiliarity of the mouth-opening movement as a neutral gesture may have resulted in a greater error signal between the predicted, usual neutral gesture the participants would expect to see, and the observed, unusual neutral gesture they were instructed to categorize as such. Additional predictions that required updating in the mouth opening condition may have activated the sensorimotor areas more than the familiar and ordinary happy and sad gestures. Future research may examine the coordinated activity of the involved brain regions by connectivity analyses to quantify the differences in their associations or dependencies under different conditions.

In conclusion, ambiguity or complexity of emotional information may result in greater activity in the sensorimotor areas if difficulty of the emotion recognition task requires a stronger engagement of the simulation system. Present findings provide support for the involvement of the MNN in face simulation, and indicate a complex relationship between sensorimotor activity and facial expression processing. Current data call for further research on the observation-related activity within and between the key brain areas involved in mimicry and social information processing. The explanations offered above which attribute the observed effect to the ambiguity of emotion may be addressed in future studies by comparing the level of activity in the premotor, motor and somatosensory areas in response to social stimuli depicting different intensities of various emotions. High spatial resolution neuroimaging techniques, such as fMRI, can be employed to investigate the involvement of the main MNN areas as well as deeper brain regions in the simulation of ambiguous motor and emotion information.

The datasets for this study will not be made publicly available because raw per-participant EEG data cannot be made publicly available as ethical clearance for sharing raw individual data on public repositories has not been received and it is not possible to do so retrospectively. Preprocessed EEG data, in a form as close to the raw data as possible, will be made available upon request of researchers.

OK performed data acquisition, analysis, interpretation of the data and wrote the first and final drafts of the manuscript. IK contributed substantially to the design of the experiment and critical revision of the manuscript. MM contributed substantially to data analysis and critical revision of the manuscript. All authors approved the final draft of the manuscript for publication (OK, MM, IK).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.