Two different mirror neuron pathways for social and non-social actions? A meta-analysis of fMRI studies

Abstract

Mirror neurons (MNs) represent a class of neurons that are activated when performing or observing the same action. Given their role in social cognition and previous research in patients with psychiatric disorders, we proposed that the human MN system (MNS) might display different pathways for social and non-social actions. To examine this hypothesis, we conducted a comprehensive meta-analysis of 174 published human functional magnetic resonance imaging studies. Our findings confirmed the proposed hypothesis. Our results demonstrated that the non-social MN pathway exhibited a more classical pattern of frontoparietal activation, whereas the social MN pathway was activated less in the parietal lobe but more in the frontal lobe, limbic lobe, and sublobar regions. Additionally, our findings revealed a modulatory role of the effector (i.e. face and hands) within this framework: some areas exhibited effector-independent activation, while others did not. This novel subdivision provides valuable theoretical support for further investigations into the neural mechanisms underlying the MNS and its related disorders.

Introduction

Mirror neurons (MNs), initially found in the premotor cortex of monkeys, are a class of neurons that can be activated when an individual performs an action or observes others performing it (Di Pellegrino et al. 1992, Gallese et al. 1996). Accumulated evidence has demonstrated the existence of a similar system, known as the mirror neuron system (MNS) or MN network, in humans (Buccino et al. 2004, Cattaneo and Rizzolatti 2009, Mukamel et al. 2010, Molenberghs et al. 2012, Keysers et al. 2018). To date, research on the MNS has expanded beyond motor actions to encompass various topics, including social and non-social cognition (Bonini et al. 2022). It is thought that the MNS partakes in these cognitive functions by recruiting somatomotor and visceromotor representations in the observer’s brain, thus aiding in the comprehension of others’ action goals and bodily states (Rizzolatti et al. 2001, Rizzolatti and Fabbri-Destro 2008). In this regard, many hypotheses concerning the mechanisms underpinning the MNS have been proposed. According to stimulationist hypotheses, such as the direct matching hypothesis (Rizzolatti et al. 2001) or the embodied simulation hypothesis (Gallese 2006), observing others’ actions or emotional expressions activates cortical mirror mechanisms, which transform sensory information about these actions and emotions into one’s own somatomotor and visceromotor representations. As such, it has been suggested that the MNS may be related to the capacity for empathy (de Waal and Preston 2017). Similarly, predictive coding accounts of the MNS focus on its potential role in predicting others’ actions and bodily states through sensorimotor predictions, positing that the MNS aids in inferring the intentions behind observed actions by minimizing prediction errors across all levels of the cortical hierarchy engaged during action observation (Kilner et al. 2007). Additionally, the Hebbian learning theory provides a potential neurobiological mechanism to explain how predictive coding is implemented at the synaptic level (Keysers and Gazzola 2014). Recently, the social affordance processing theory has proposed a more extensive role for the MNS in a broader spectrum of social interactions (Orban et al. 2021a, 2021b). This theory suggests that the MNS may enhance an individual’s ability to anticipate and prepare for potential actions in social contexts, thereby supporting social cognition and interactive behaviors. The above hypotheses collectively suggest that mirror mechanisms like the human MNS may represent a fundamental principle of brain function (Rizzolatti and Sinigaglia 2016, Bonini et al. 2022). Therefore, the discovery of MNs has been considered one of the most remarkable breakthroughs in neuroscience (Keysers and Fadiga 2008).

To gain a more comprehensive understanding of the MNS’s organization, many studies have attempted to analyze its subdivisions. For example, Cattaneo et al. partitioned the MNS based on types of actions, such as transitive distal movements, reaching movements, and tool-use movements (Cattaneo and Rizzolatti 2009). Furthermore, several meta-analyses explored the differences among MN pathways associated with observation, imitation, imagination, execution, etc., revealing a similarity among these networks (Molenberghs et al. 2009, Van Overwalle and Baetens 2009, Caspers et al. 2010, Hardwick et al. 2018; Savaki and Raos 2019). For instance, action observation and execution recruit similar bilateral premotor, parietal, and sensorimotor networks. However, bilateral sensorimotor areas are more consistently associated with action execution, while action observation recruits mainly bilateral premotor, parietal, and occipital areas in comparison to action execution (Hardwick et al. 2018). In addition, action observation and imitation share a bilateral network involving the frontal premotor, parietal, and occipitotemporal cortices while exhibiting specific differences in Broca’s area (Caspers et al. 2010).

Recently, it has been proposed that the fundamental responsiveness of MNs to others’ bodily displays may constitute the basis of social cognition in primates (Rizzolatti and Fabbri-Destro 2008, Schmidt et al. 2021), whose intricate social environments have exerted substantial cognitive demands, shaping the evolution of their brains (Byrne and Whiten 1988, Dunbar 1998). Note that humans employ a diverse array of cues, such as facial expressions, speech, and social hand gestures, to communicate efficiently, interact effectively, and accurately gauge the intentions of those around them (Saggar et al. 2014). As proposed by the social affordance processing theory, the MNS may play a crucial role in managing online social interactions (Orban et al. 2021a, 2021b). Therefore, this raises the question of whether the MNS processes social and non-social actions differently. This potential division within the MNS has been hinted at by findings from patients with social dysfunction disorders such as autism spectrum disorder (ASD), where impairments in the MNS may account for key symptoms (Ramachandran and Oberman 2006). Several systematic reviews have concluded that studies employing social stimuli (e.g. facial expressions and whole-body fearful actions) have exhibited altered activation of the MNS in ASD patients compared to controls, whereas studies utilizing non-social action stimuli (e.g. finger movements) have not (Hamilton 2013, Kilroy et al. 2019, Chan and Han 2020). On the contrary, although the patients with Williams syndrome have an intellectual disability and are severely impaired in visuospatial function, they show an exaggerated interest in other people and remarkable expressiveness and social communication abilities (Bellugi and George 2000, Adolphs 2003, Sparaci et al. 2012), which may also be associated with the dysfunction of the MNS (Järvinen et al. 2013, Ng et al. 2016). Therefore, understanding such divisions within the MNS is crucial for comprehending its organization and implications in social dysfunction disorders like ASD, which may, in turn, pave the way for developing highly effective therapeutic interventions.

So far, empirical evidence remains scarce regarding whether the MNS exhibits different pathways for social and non-social actions. Recently, it has been proposed that the MNS in monkeys comprises two different anatomical pathways for facial and hand actions, challenging the conventional view that both types of actions share the same neuroanatomical network (Ferrari et al. 2017). Note that the research on MNs in monkeys is restricted to a limited range of actions. In particular, almost all hand actions studied are non-social (e.g. reaching and grasping movements) (Casile 2013). Therefore, it remains unclear whether these two pathways in monkeys reflect a division within the MNS between facial and hand actions or, indeed, between the social information carried by facial actions and the non-social information carried by hand actions. However, such clarification in monkeys is challenging due to the limited variety of action types available, particularly the absence of hand actions with social meanings. In humans, several recent reviews and meta-analyses have demonstrated that the MNS for facial expressions and emotions includes the amygdala, insula, and anterior cingulate gyrus (ACC), in addition to the classical frontoparietal regions, and thus suggested the presence of different networks for motor and emotional stimuli (Rizzolatti and Sinigaglia 2016, Bonini et al. 2022, Del Vecchio et al. 2024). Emotions play a crucial role in social actions, but they are not the only important aspects. Other facial actions, such as speech, also convey social information. Furthermore, human actions are diverse, encompassing social actions not only from the face but also from the hands. Hand actions, such as expressive gestures (e.g. thumbs-up and OK) and actions involving interpersonal interaction (e.g. shaking hands), also convey social information. Similarly, non-social actions also include activities involving the face in addition to hands, such as biting and eating. The wide range of actions in humans provides us with the opportunity to investigate the potential division of the MNS along the dimensions of social and non-social actions.

In the present study, we aimed to systematically explore whether the human MNS exhibits different pathways for social and non-social actions and the potential modulatory role of the effector (i.e. face and hands) on sociality. We reviewed published functional magnetic resonance imaging (fMRI) studies on the MNS, categorized the actions, and conducted meta-analyses to explore the potential subnetworks within the MNS.

Materials and methods

Literature selection

This meta-analysis was performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (Moher et al. 2009, Page et al. 2021a, 2021b). fMRI studies included in the present meta-analysis were obtained from the PubMed and Web of Science databases (as of June 2023). Considering that most previous studies did not explicitly differentiate between social and non-social stimuli, we initiated our search by focusing on face- and hand-related literature. Subsequently, we recategorized these studies based on the social nature of the stimuli to better clarify the distinction between social and non-social actions. For face literature, the search string employed was “(mirror AND fMRI) AND (face OR facial OR lip OR mouth OR eye).” Similarly, the search string used for hand-related literature was “(mirror AND fMRI) AND hand.” To ensure comprehensive coverage, we also referred to action observation studies listed in the earlier meta-analysis by Hardwick et al. (2018). To extend their search, we included studies published between June 2017 (the cutoff date of the studies included in Hardwick et al. 2018) and June 2023, utilizing the exact keyword string “fMRI AND (Action Observation OR mirror neurons OR imitation).” An overview of the literature search and study screening are shown in Fig. 1. Finally, we screened a total of 48 face-related and 126 hand-related papers according to the following criteria:

1. Studies employing fMRI as the imaging technique were included, while those utilizing other imaging modalities such as positron emission tomography (PET), magnetoencephalography, and electroencephalography were excluded. It is well known that different imaging modalities can influence activation profiles due to methodological differences (Costafreda 2009, Hardwick et al. 2015). Consequently, the various components of fMRI and PET studies across conditions may contribute to the observed differences between these conditions. Therefore, we restricted our study to fMRI data to ensure approximately comparable spatial and temporal resolution for the activation likelihood estimation (ALE) analyses.

2. Reviews and meta-analyses were excluded.

3. Only studies mentioning MNs were included to align with our study topic and avoid ambiguity.

4. Only studies conducted on healthy adult subjects were included. Experiments involving macaques, children, older adults, and patients were excluded. Note that data from healthy controls in patient studies were still considered. Furthermore, studies involving specialized professional subjects (e.g. experiments that recruited musicians to observe the MNS associated with piano-playing actions) were also excluded to eliminate the influence of expertise.

5. Studies reporting concurrent activation of observation and execution/imitation were included. Additionally, studies only involving action observation were also included. However, studies only involving action execution or imitation were excluded to adhere to the more stringent definition of MNs. In addition, we did not include second-person interactions, where social signals are exchanged between two interacting agents (Schilbach et al. 2013; Wang et al. 2018, Redcay and Schilbach 2019, Freiwald 2020). Previous studies have shown that second-person interactions significantly modulate the MNS and activate the mentalizing network. In the present study, we primarily focused on the conventional MNS. To limit the potential impact of mentalizing functional networks on the MNS, we therefore excluded second-person interactions.

6. Studies utilizing abiotic or nonvisual stimuli, such as robot actions, point-light figures, animation figures, auditory stimuli, and tactile stimuli, were excluded.

7. Studies that reused existing data or did not report coordinates were excluded.

8. Only studies including coordinates from whole-brain analysis in standard MNI or Talairach space were included to avoid inflated significance for specific regions stemming from over-represented region of interest analysis (Muller et al. 2018).

During the literature search, screening, and data collection, three experimenters cross-checked the results.

Literature classification and foci selection

The selection of literature for the meta-analysis involved the collection of activation coordinates, the number of participants, and other information from the chosen 48 face (Face-All) and 126 hand-related (Hand-All) studies. Details of individual experiments are shown in Supplementary Table S1. The reported Talairach space coordinates were transformed into the MNI space using the icbm2tal algorithm (Lancaster et al. 2007) implemented in the Ginger ALE software (Eickhoff et al. 2009). In total, 958 face-related and 2366 hand-related activation coordinates (foci) were included in the meta-analysis.

To explore the potential pathways associated with sociality, we divided the screened literature into social (Social-All) and non-social (Non-Social-All) actions. Social actions encompass a wide range of behaviors such as emotional expressions (e.g. happy and angry), speech, various gestures that convey information (e.g. thumbs up and ‘give it to me’), and actions involving interpersonal interaction (e.g. shaking hands). Non-social actions, on the other hand, refer to behaviors that do not convey social information to others, including actions lacking social significance (e.g. finger tapping and mouth opening) and actions related to interacting with objects (e.g. biting and grasping).

Furthermore, to examine the potential modulatory role of the effector (i.e. face and hands) on sociality, the selected literature was further divided into four subcategories: Face-Social (FS), Hand-Social (HS), Face-Non-Social (FNS), and Hand-Non-Social (HNS). We classified social and non-social actions based on the classifications provided by the original studies. Where classifications were unclear in the original studies, three experimenters reached a consensus based on descriptions within the original studies. If no clear classification could be made, we categorized these actions as “All” but did not classify them as social or non-social. The number of papers included in each subcategory and category is presented in Table 1, and the specific action types within each subcategory are listed in Table 2. Note that the total number of papers in the Hand-All category does not equal the sum of the literature in the HS and HNS subcategories due to certain studies employing both social and non-social conditions.

ALE meta-analysis

Analyses were conducted to identify MN pathways associated with Social-All, Non-Social-All, Face-All, and Hand-All conditions. Additionally, networks linked to FS, HS, FNS, and HNS actions were explored. The meta-analysis employed the updated version of the ALE algorithm, as implemented in GingerALE (version 3.0.2) (Eickhoff et al. 2009, 2011, 2012, Turkeltaub et al. 2012). This algorithm determines whether converging foci across different studies occur at a level greater than the null distribution. Modeled activation maps were created using reported foci and Gaussian blur with an full-width at half-maximum empirically derived from sample size. Studies with larger sample sizes were modeled with narrower Gaussian distributions (Eickhoff et al. 2009). ALE scores, which describe the convergence of coordinates for each location, were calculated based on the union of modeled activation maps across studies. The null distribution of the ALE statistic was obtained using label-exchangeability to generate a P-value table. The threshold for ALE scores was set at P < .05 [cluster-level family-wise error (FWE) corrected for multiple comparisons, 1000 threshold permutations, with a cluster-forming threshold at the voxel level of P < .001] according to the previous studies (Eickhoff et al. 2012, 2016). The resulting ALE map for each category or subcategory was provided at a voxel resolution of 2 mm³.

Subsequently, ALE contrast analyses were performed to identify conjunct and specific clusters between the compared ALE maps (Eickhoff et al. 2012). We generated the contrast image by comparing the two ALE maps. Statistical significance was determined by comparing it with an empirical null distribution. To account for differences in study size between the compared categories/subcategories, the null distribution was constructed using GingerALE by pooling datasets from the two compared categories/subcategories and randomly dividing them into two new groups with the same size as the original datasets. Voxel-wise differences between these two maps were calculated. This permutation procedure was iterated 1000 times to obtain a sufficient sample of ALE null distributions. In GingerALE, the thresholded ALE maps from individual analyses were used to calculate conjunction and contrast analyses using the conservative minimum statistic (Nichols et al. 2005) to identify significant voxels (Alain et al. 2018, Aziz-Safaie et al. 2024). Notable, cluster-level FWE correction for conjunction and contrast analyses is unavailable in GingerALE (Hoffman and Morcom 2018, Papitto et al. 2020, Arioli et al. 2022). Consequently, we adopted a conventional uncorrected P < .05 threshold and a minimum cluster volume of 100 mm³, according to previous publications (Alain et al. 2018, Hardwick et al. 2018, Papitto et al. 2020, Arioli et al. 2022, Gan et al. 2022, Aziz-Safaie et al. 2024).

Analysis of network similarity

To quantitatively evaluate the similarity between the two ALE maps, the Dice coefficient was calculated. The Dice coefficient measures the extent of overlap between two maps and is computed as shown in Equation (1):

where |V1 ∩ V2| represents the number of overlapping voxels between ALE maps 1 and 2, while |V1| and |V2| represent the number of voxels in each ALE map, respectively. A Dice coefficient value of 1.0 indicates complete overlap, while a value of 0.0 indicates no overlap.

To ensure accurate calculation of the Dice coefficient, the two comparable maps needed to be of similar sizes. As such, we randomly selected the same number of voxels with replacements from both ALE maps (n = the number of voxels in the smaller map). The Dice coefficient was then calculated for the resampled maps. This resampling procedure was repeated 1000 times to estimate the distribution of the Dice coefficient. Next, the resulting distribution was compared with the chance level (0.5) to determine the statistical significance of the Dice coefficient between the two ALE maps (Guindon and Zhang 2016).

Studies resampling analysis

Given the variation in the number of studies across categories/subcategories, several resampling analyses were conducted to validate the main findings.

Unlike contrast analysis, conjunction analysis does not account for differences in sample sizes across categories. Therefore, to create a fair conjunction map between the Social-All and Non-Social-All ALE maps, 53 articles (the smaller number of articles in the two categories, i.e. Social-All) were randomly selected without replacement from the category with more articles (i.e. Non-Social-All) to create a Non-Social-All dataset matched with the Social-All one and the corresponding ALE map. Then, a resampling conjunction map was created. The Dice coefficient [averaged across 1000 times of voxel resampling (see details in Section Analysis of network similarity)] between the original and resampling conjunction maps was then calculated. This resampling process was repeated 1000 times to determine the statistical significance.

Notably, there was a degree of overlap in the literature between the Social-All and Face-All categories, as well as between the Non-Social-All and Hand-All categories. To eliminate the potential impact of this overlap on our findings, we balanced the proportions within each subcategory. That is, 10 articles (the minimum number of articles across the four subcategories, n = 10 in FNS) were randomly selected without replacement from the FS, HS, and HNS subcategories. These selected articles were then used to create well-balanced datasets for Social-All, Non-Social-All, Face-All, and Hand-All. Subsequently, we calculated Dice coefficients among different ALE maps. Again, this resampling process was repeated 1000 times to determine the statistical significance.

Presentation of results

The visualization and anatomical labeling of the results were performed using Mango (version 4.1, downloaded from http://brainmap.org) software. The results were overlaid on a standard brain template in the MNI space. Anatomical labeling was conducted by referring to the nearest gray matter MNI labels provided by Mango.

Fail-Safe N analysis

To account for the file drawer problem, which is a type of publication bias that refers to unpublished experiments and can potentially influence results and the validity of findings (Hayasaka et al. 2018, Samartsidis et al. 2020), the Fail-Safe N (FSN) analysis was performed. Previous studies have estimated a 95% confidence interval for the number of studies that report no local maxima varies from 5 to 30 per 100 published studies (Hayasaka et al. 2018, Samartsidis et al. 2020). To test the robustness of the current meta-analyses, we applied the upper bound of the estimated maximum number of null studies in the file drawer by introducing the 30% noise (the maximum number of null studies estimated to be in the file drawer) for each category (i.e. Social-All, Non-Social-All, Face-All, Hand-All, FS, HS, FNS, and HNS).

Results

MN pathways for Social-All and Non-Social-All

Initially, we conducted meta-analyses to explore the MN pathways associated with Social-All and Non-Social-All conditions.

The meta-analysis revealed that Social-All recruited a widespread network spanning bilateral frontal, right parietal, bilateral temporal, bilateral occipital, and bilateral limbic lobes (Fig. 2a). In the frontal lobe, the activation elicited by social actions involved the bilateral inferior frontal gyrus (IFG), bilateral middle frontal gyrus (MFG), right superior frontal gyrus (SFG), and right precentral gyrus (PrG). In the parietal lobe, social actions evoked activation in a small portion of the right inferior parietal lobe (IPL). Moreover, social actions also activated the temporal and occipital areas, including the left inferior temporal gyrus (ITG), right middle temporal gyrus (MTG), bilateral superior temporal gyrus (STG), bilateral fusiform gyrus (FuG), right inferior occipital gyrus (IOG), right middle occipital gyrus (MOG), and left lingual gyrus. Notably, the MN pathway associated with the Social-All condition also recruited the bilateral amygdala.

Non-social actions activated frontal areas, including the right IFG, right MFG, right SFG, left medial frontal gyrus (MeFG), and bilateral PrG (Fig. 2a). In response to non-social actions, activated clusters were observed in the parietal lobe, including bilateral IPL, left superior parietal lobule (SPL), bilateral postcentral gyrus (PoG), and bilateral precuneus (Pcun). In addition, the MN pathway for Non-Social-All recruited similar areas within bilateral occipital and temporal lobes as the MN pathway for Social-All (Supplementary Table S2). Only the bilateral insula, but not the amygdala, was activated by non-social actions. Notably, although both social and non-social actions activated the right IFG, social-induced activation was more anterior and lateral.

To quantify the difference between MN pathways for Social-All and Non-Social-All, we calculated the Dice coefficient, which yielded a value of 0.318 (P = 1), indicating limited similarity. To further explore the detailed similarities and differences between MN pathways for Social-All and Non-Social-All, we conducted ALE contrast analyses to identify overlapping and specific regions for MN pathways associated with social and non-social actions.

The conjunction maps revealed that both social and non-social actions evoked responses in the bilateral IFG, left SFG, right PrG, right IPL, left ITG, bilateral MTG, right STG, bilateral FuG, right IOG, and right MOG (Table 3). After accounting for differences in sample sizes between Social-All and Non-Social-All categories by conducting the resampling analysis, we obtained similar results: the mean Dice coefficient between the resampling and the original conjunction maps across 1000 iterations was 0.776 (P < .001).

In the frontal lobe, social actions evoked stronger activation in bilateral IFG, bilateral MFG, right SFG, right MeFG, and right PrG (Fig. 3a). In contrast, non-social actions more strongly activated the right MFG, left MeFG, and bilateral PrG. Notably, the activated cluster in the right PrG associated with social actions was more anterior, lateral, and inferior, whereas that related to non-social actions was more posterior, medial, and superior. In the right MFG, the activation evoked by social actions was more posterior, whereas the activation elicited by non-social actions was more anterior. Moreover, in the parietal lobe, no region was more activated during social than non-social actions, whereas non-social actions evoked stronger activation in the bilateral IPL, bilateral PoG, and bilateral Pcun (Fig. 3b). Additionally, social actions evoked stronger activation in the bilateral amygdala and right insula than non-social actions.

The modulatory role of the effector (i.e. face and hand) on MN pathways for Social-All and Non-Social-All

MN pathways for Face-All and Hand-All

To investigate the impact of the effector (i.e. face and hands) on the MNS organization along the sociality (social and non-social) dimension, we first investigated MN pathways associated with Face-All and Hand-All conditions. As shown in Fig. 2b, both facial and hand actions elicited MN pathways that exhibited striking similarities to those observed for social and non-social actions, respectively. Similar to social actions, facial actions recruited a network including areas within frontal (i.e. bilateral IFG, right SFG, and right PrG), parietal (i.e. right IPL), temporal (i.e. left ITG, right MTG, bilateral STG, and bilateral FuG), and occipital lobes (i.e. right IOG), as well as the bilateral amygdala. On the other hand, hand actions activated a network akin to the MN pathway of Non-Social-All, involving areas spanning frontal (i.e. right IFG, right MFG, right SFG, left MeFG, and bilateral PrG), parietal (i.e. bilateral IPL, left SPL, left PoG, and bilateral Pcun), temporal (i.e. bilateral ITG, left MTG, and right FuG), and occipital lobes (i.e. left IOG and right Lingual Gyrus), as well as bilateral insula (Fig. 2b).

The similarity between MN pathways for Social-All and Face-All, as well as between MN pathways for Non-Social-All and Hand-All

To quantify the similarity between MN pathways for Social-All and Face-All, as well as between MN pathways for Non-Social-All and Hand-All, we employed Dice coefficient analyses. The Dice coefficient analyses yielded a value of 0.814 (P < .001) for the similarity between MN pathways for Social-All and Face-All and 0.911 (P < .001) for the similarity between MN pathways for Non-Social-All and Hand-All. These values indicated substantial overlap between MN pathways of Social-All and Face-All and between MN pathways of Non-Social-All and Hand-All. These findings were further corroborated by additional ALE contrast analyses (Fig. 4), which revealed extensive overlap between MN pathways for Social-All and Face-All but no significant specific regions associated with either condition. Similar findings were observed when comparing MN pathways for Non-Social-All and Hand-All conditions.

Notably, the literature in the Social-All and Face-All categories overlapped to a certain degree, as did the literature in the Non-Social-All and Hand-All categories. To eliminate the potential impact of such overlap on the above findings, we conducted a resampling analysis (n = 1000) to balance the proportions of literature in each category (i.e. 10 articles for each of the four subcategories: FS, HS, FNS, and HNS). The resampling analysis yielded a moderate similarity between MN pathways for Social-All and Face-All (mean Dice index = 0.486, P = .499) as well as between MN pathways for Non-Social-All and Hand-All (mean Dice index = 0.564, P = .227). The lack of significance could be attributed to the reduced number of articles after resampling (n = 20 for each category) compared to the original analysis. Note that the Dice coefficient between MN pathways for Non-Social-All and Hand-All was significantly higher than that between Social-All and Hand-All (P = .025). Similarly, the Dice coefficient between MN pathways for Social-All and Face-All was marginally significantly higher than that between Non-Social-All and Face-All (P = .065). These results indicate that the MN pathway activated by social actions may be more similar to that activated by facial actions than by hand actions, whereas the MN pathways activated by non-social actions are more similar to those activated by hand actions, even after accounting for the proportion of literature.

Sub-MN pathways for FS, HS, FNS, and HNS

To further investigate the impact of the effector (i.e. face and hands) on the MNS organization along the sociality (social and non-social) dimension, we conducted meta-analyses for each subcategory. All four subcategories evoked activations in the occipitotemporal lobe, albeit with some variations (Supplementary Table S2). Notably, there were several differences among the sub-MN pathways in the frontal, parietal, and limbic lobes, as well as sublobar regions. FS actions elicited activations in bilateral IFG, right SFG, right PrG, bilateral amygdala, and left cingulate gyrus (CG) (Fig. 5a). HS actions engaged a frontoparietal network comprising the left IFG, MFG, and PrG (Fig. 5a). FNS actions did not recruit any region other than the occipitotemporal lobe (Fig. 5b). HNS actions evoked widespread activation in the frontal and parietal lobes, including right IFG, right MFG, right SFG, left MeFG, bilateral PrG, bilateral IPL, left SPL, bilateral PoG, and bilateral Pcun (Fig. 5b).

Differences between MN pathways associated with social and non-social actions that are dependent and independent of effectors

Next, we analyzed differences between MN pathways associated with social and non-social actions that are dependent and independent of effectors, particularly the latter. To achieve this, we compared MN pathways associated with FS and FNS actions, as well as MN pathways associated with HS and HNS actions, respectively. Suppose a region is more strongly activated by FS actions than by FNS ones and also more strongly activated by HS actions than by HNS ones. In that case, this overlapping region is defined as associated with social action independent of the effector. The same principles were applied to the non-social MN pathway.

Although there were regions that showed stronger activation for FS compared to FNS, as well as for HS compared to HNS, no effector-independent regions were found for social actions or non-social actions (Supplementary Fig. S1). This result may be due to the limited number of articles on FNS and HS actions.

To mitigate the impact of the limited number of articles, we used an uncorrected cluster-forming threshold (P < .001) without the cluster-level FWE correction (Hoffman and Morcom 2018) in the individual analyses for FNS and HS. To ensure fair comparisons across subnetworks, we also applied the same uncorrected cluster-forming threshold for FS and HNS actions. As shown in Table 4, we observed regions associated with social actions (e.g. right IFG, left amygdala, and left insula; Fig. 6a and b) and non-social ones (e.g. left PrG, left PoG, and left thalamus; Fig. 6c and d) independent of the effector. Although the left parahippocampal gyrus (PHG) was involved in both social and non-social actions, its spatial location differed: the left PHG associated with social actions was more medially located, whereas the left PHG associated with non-social actions was more laterally positioned (Supplementary Fig. S2). There were also regions specific to social and non-social actions depending on the effector. Details are provided in Table 4.

FSN analyses

Finally, FSN analyses were conducted to assess the robustness of the findings mentioned earlier. The results revealed sufficient robustness, with only the two smallest clusters in the MN pathway for Social-All actions [around the left IFG and MFG, near Brodmann area (BA) 9; 25% < FSN < 30%] and the one for HNS actions (around the Cuneus in the right occipital lobe, BA 7; 5% < FSN < 10%) having FSNs <30%. Note that another cluster around area BA 45 in the left IFG in the Social-All MN pathway survived even after adding 30% noise. These findings indicate an overall robust convergence of foci found in the present study.

Discussion

In the present study, we conducted meta-analyses to examine the potential division of the MNS along the dimension of social and non-social actions and the impact of the effector (i.e. face and hands) on it. Our results showed the existence of two different MN pathways for social and non-social actions, with certain brain regions being independent of the effector (Fig. 7) and others not. Below, we discuss the importance of these findings for understanding the MNS.

Different MN pathways for social and non-social actions

Both social and non-social actions evoked activations in the bilateral IFG, left SFG, right PrG, and right IPL, along with activations in areas within bilateral temporal and occipital lobes. Note that social actions recruited more areas in the frontal and limbic lobes and sublobar regions, whereas non-social actions were more associated with the parietal lobe. These results suggest that there may exist two different MN pathways for social and non-social actions. Non-social actions activated the classical MN pathway (e.g. bilateral parietal areas IPL, PoG, and Pcun, as well as frontal areas right MFG and bilateral PrG), in which information is transmitted from the occipitotemporal lobe to the parietal lobe and then to the frontal lobe (Nelissen et al. 2011). However, the social MN pathway included fewer parietal regions instead of more frontal, limbic, and sublobar regions, including bilateral IFG, right SFG, right PrG, bilateral amygdala, and right insula. Notably, in the frontal lobe, activations evoked by social actions were located more anteriorly and laterally than non-social actions. Moreover, our findings revealed the involvement of the amygdala, which plays an important role in social cognition (Adolphs 1999, 2010), in the social MN pathway but not in the non-social MN pathway.

We also explored the impact of the effector (i.e. face and hands) on the social and non-social MN pathways by subdividing the actions into four distinct subcategories: FS, HS, FNS, and HNS. We observed slightly different patterns in the frontal lobe, parietal lobe, limbic lobe, and sublobar regions among the four sub-MN pathways (Fig. 5). Through a series of contrast analyses, we identified such subtle differences by controlling for the effector factor and then comparing sociality at different levels (e.g. FS versus FNS). Our results indicated that the right IFG, left amygdala, and left insula were more related to social than non-social actions independent of the effector (Fig. 6, Table 4). Conversely, the left PrG, PoG, and thalamus were more involved in non-social than social actions, regardless of the effector (Fig. 6, Table 4). There were also brain areas associated with social or non-social actions depending on the specific effectors (Table 4). Therefore, our findings suggest that there may exist different social and non-social MN pathways, with effectors (i.e. face and hands) playing a modulatory role.

Previous studies have established an emotional versus motor framework for the MNS, wherein the amygdala, insula, and ACC are more associated with emotions, while the MN pathway associated with motor activities (e.g. grasping and manipulation) involves parieto-frontal regions such as the premotor cortex and IPL (Rizzolatti and Sinigaglia 2016, Bonini et al. 2022, Del Vecchio et al. 2024). Given that emotions are a primary component of social actions (especially face-related ones) and motor activities are central to non-social actions (especially hand-related ones), it is unsurprising that the social and non-social MN pathways identified in the present study overlap to some extent with the emotional versus motor framework. However, it is noteworthy that we found that most emotion-related MNS regions, including the amygdala and insula, were also more activated by HS actions than by HNS ones (Fig. 6, Table 4). Our results showed that the CG, which is located near/overlapped the ACC as reported in the emotional framework, was more related to FS actions rather than HS actions (Fig. 5). Similarly, we found that the frontal (e.g. left PrG) and parietal (e.g. left PoG) regions in the MNS associated with motor activities were indeed more involved in HNS actions than in HS ones (Fig. 6, Table 4). Notably, these regions were also more involved in FNS actions than in FS actions. Therefore, our findings indicate that it may be more appropriate to divide the MNS along the dimension of sociality rather than the emotional versus motor framework.

The two different MN pathways for social and non-social actions found in the present study not only deepen our understanding of the organization of the MNS but also explain the dysfunctions observed in individuals with certain disorders (e.g. ASD and Williams syndrome). For example, our findings may provide neural circuitry evidence for current ASD models: the emulation and planning-mimicry (EP-M) model (Hamilton 2008), which describes an indirect parietal route for goal emulation and planning, and a direct occipital-frontal route for mimicry, as well as the social top-down response modulation (STORM) model (Hamilton 2013; Wang and Hamilton 2012). According to our findings, the social difficulties in patients with ASD may mainly arise from dysfunction within the social MN pathway (e.g. IFG, SFG, and amygdala). Such dysfunction may explain the dysfunction of the MN pathway in the EP-M model and the top-down regulation of MNS in social contexts in the STORM model. Note that the present meta-analysis focused exclusively on healthy adults. Future research conducting a comparison between the social and non-social MN pathways in individuals with ASD may provide more direct evidence to further enhance our understanding of the neural mechanisms underlying ASD.

The similarity between MN pathways for Social-All and Face-All, as well as between Non-Social-All and Hand-All

In the present study, we found that the MN pathway activated by social actions may be more similar to that activated by facial actions than by hand actions, whereas the MN pathway activated by non-social actions is more similar to those activated by hand actions, even after accounting for the proportion of literature. The similarity between social and face MN pathways, as well as non-social and hand MN pathways, may be linked to their development models. During development, motor commands responsible for controlling hand movements are usually integrated with visually tracking their own hand movements in space, guiding infants in reaching and grasping objects. Therefore, the emergence of visuomotor associations for non-social hand actions may occur at an earlier stage in development. These associations are essential in establishing visuomotor information integration within the parietal–premotor network (Del Giudice et al. 2009, Casile et al. 2011, Ferrari et al. 2013, 2017, Cook et al. 2014, Tramacere and Ferrari 2016). Conversely, since infants cannot observe their own facial actions directly and cannot recognize their own faces in the mirror until ∼18 months (Amsterdam 1972), the face MN pathway may primarily develop through interactions, predominantly social interactions, between infants and caregivers. As such, the development of face MNs may be closely linked to social actions (Ferrari et al. 2013, 2017, Murray et al. 2016).

Limitations

Despite our efforts to gather a wider range of action types in this study, we still encountered an imbalance in the selected literature due to the limitations of current research. First, FS actions were predominantly represented by facial expressions, while HNS actions primarily consisted of manipulative and/or object-oriented hand actions. Consequently, these two major action types may dominate our results concerning the MN pathways for FS and HNS actions. Indeed, it has been established that facial expressions inherently activate the limbic lobes (Bonini et al. 2022, Del Vecchio et al. 2024). Moreover, previous evidence of structural connectivity indicates that the limbic system is linked to both the ventral premotor area and executive control areas (Augustine 1996, Ghashghaei and Barbas 2002). In contrast, manipulative and/or object-oriented hand movements inherently activate parietal nodes (Cerliani et al. 2022; Orban et al. 2021b). Additionally, HS and FNS actions were more diverse but with fewer sample sizes than FS and HNS actions. Furthermore, in our research, the majority of included studies were observational. Only a small portion involved participants both observing and imitating or executing actions. Moreover, we did not include second-person interactions to eliminate the potential impact of the mentalizing network (see Materials and methods). To better address these issues, further investigation with more diverse studies may be necessary.

On the other hand, FS actions, such as emotional expressions, can be precisely abstracted as social actions, while such abstraction is challenging for hand gestures without context (e.g. grasping). The potential mismatch between participants’ understanding and experimenters’ classifications may cause possible issues in data interpretation. Future research with clearer contextual frameworks for hand actions may be necessary to further examine our findings regarding MN pathways associated with HS and HNS actions.

Conclusions

In summary, our study provided a systematic meta-analysis based on previous fMRI studies to differentiate MNS for social and non-social actions for the first time. We demonstrated the existence of two different MN pathways for social and non-social actions, with effectors (i.e. face and hands) playing a modulatory role. This organization expands the existing framework that emotional and motor stimuli engage different mirror pathways. Our findings provide novel insights into the hierarchical organization of the MNS and offer valuable guidance for future research on the MNS and the exploration of neural mechanisms underlying disorders related to the MNS (e.g. ASD or Williams syndrome).

Supplementary data

Supplementary data is available at SCAN online.

Conflict of interest

None declared.

Funding

This study was supported by the Ministry of Science and Technology of China STI2030-Major Projects (Nos. 2021ZD0204200 and 2021ZD0200200), the National Natural Science Foundation of China (No. 32071094).

Data availability

Data will be made available on request.

References