The mirror neuron system (MNS), which is a physiological substrate that may subserve certain mechanisms underlying social cognition has recently gained a lot of attention from the research community. MNS is a system consisting of visuomotor neurons that are responsible for the mirror mechanisms, meaning that each time an individual observes another individual performing an action, these neurons which encode that action, are activated in the observer's cortical motor system (1). Observed activations of this system are referred to as mirror neuron activity (MNA). MNA is considered a subdomain of social cognition (2). Several important functions beyond the action domain have been theorized for MNS, such as being a fundamental building block for understanding others' actions (3), encoding the intentions of the actor (4, 5), facilitating imitation (6, 7), and playing a role in human infants' ability to map similarities between self and others (8). Additionally, there has been an emphasis on the possible ties between MNA and empathy (9), and language (10). Mirror neurons were first discovered in the premotor area F5 of macaque monkeys (11). Later, similar neurons were found in the inferior parietal lobule, area PF, of macaque monkeys, and the concept of MNS was established (12). Some studies have claimed the discovery of similar neurons in various regions of the human brain, including the ventral premotor cortex (PMv) (13, 14), inferior frontal gyrus (IFG) (15–17), superior temporal sulcus (STS) (18–20), and inferior parietal lobule (IPL) (14, 21). In the meantime, some counter-arguments exist that question the function and even the very existence of the human MNS, with the strongest argument being the absence of single-cell recording data for human subjects (22). These counter-arguments were assuaged following single-cell recording studies in pre-surgical patients (23), the repetition suppression functional magnetic resonance imaging (fMRI) procedure in healthy volunteers (17), and lesion study in the human brain regions that have been proposed to be associated with human MNS (24). Nevertheless, MNS has been one of the most widely investigated domains of social cognition in psychiatric disorders within human beings. Even a recent paper by Heyes and Catmur (25) has called for more research on this phenomenon.

Other regions have also been proposed as an extension to MNS, one of the most important of them being the Brodmann area 2 (BA2) (26), which is the strongest generator of the mu rhythm (27). Mu rhythm (oscillations from 8 to 13 Hz) suppression has been proposed to be an indication of the MNA, as it is seen both when an individual performs and observes an action (28–30). A meta-analysis has demonstrated that there might also be other brain regions that do not have mirror properties but may convey necessary information to MNS including the primary visual cortex, supplementary motor area, dorsal premotor cortex, superior parietal lobe, cerebellum, and parts of the limbic system (31).

In the context of psychological disorders, MNA has been mostly investigated in autism. This is due to the “broken mirror” hypothesis and its role in explaining the social and language deficits of this disorder (32). However, research has produced insufficient evidence to support this hypothesis in its pure form, and instead, two alternative models have been proposed: the EP-M model and the social top-down response modulation (STORM) model. The STORM model proposes that autism symptoms originate from abnormalities within the top-down regulation of the MNS, rather than within the MNS itself, while the EP-M model proposes that imitation behavior in autistic individuals is served by the pathways between brain areas associated with MNS (33). Nevertheless, both these alternative models also suggest that there might be some possible dysfunction in the MNA in these patients, either within the MNS itself or within the systems that regulate MNA (32, 33). The discovery of such dysfunction has opened the doors to investigate and explain the underlying pathophysiology of the disorder in a more elaborate way and to come up with new rehabilitation methods (34–36).

Schizophrenia is one of the most debilitating and common neuropsychiatric disorders, with an estimated prevalence between 0.28 and 0.75% in the population worldwide (37–40). Deficits in a variety of cognitive domains are well-known for this disorder, such as impaired attention, verbal memory, and social cognition, and these are listed as specifiers for schizophrenia in the 11th revision of the International Classification of Diseases (ICD-11) (41). There are several reports of individuals with both autism and schizophrenia (42–45), which reveal that deficits in the theory of mind (ToM) exist in both disorders. Also, there are reports that both disorders share several genetic signals (46). A previous meta-analysis (47) of fMRI studies on autism and schizophrenia patients during ToM tasks revealed hypoactivation in the STS area, one of the main brain regions associated with MNA, in both groups, yet again emphasizing the deficits in ToM in both disorders, and possibly, hypothesizing the presence of MNA impairments in schizophrenia similar to the already known MNA impairments in autism patients.

MNA dysfunction might be another commonality between these disorders. Investigations of MNA in schizophrenia have not been as profound and conclusive as in autism. Based on a recent review (48) that partly examined this subject, findings of the state of MNA function in schizophrenia are mixed, with some studies suggesting impaired MNA function in the patients, while others did not find such a phenomenon. If such dysfunction is proven to be present in schizophrenia patients, it might potentially serve for implementing new rehabilitation treatments based on MNA training, as such treatment options have been previously found in multiple reports to be beneficial for autism patients (49–53).

As we included data from multiple paradigms, with different outcomes, quantitative analysis was not feasible. Thus, we aimed for qualitative analysis. The full texts of the included studies were read and evaluated by AV and MMo. We determined the direction of the effect based on the studies' results, as either “decreased,” “intact,” or “increased MNA” for the primary outcome and “intact” or “altered MNS” for the secondary outcome. Regarding the primary outcome, we only included studies that have directly evaluated MNA. We did not consider studies that assessed other cognitive domains hypothesized to be related to MNA (e.g., empathy, etc.) as eligible for analysis. For the secondary outcome, we did not perform any analysis as there were very few studies for this purpose.

AV analyzed the data using Microsoft Excel and dmetar (68) package for R version 4. A qualitative meta-analysis was performed for the primary outcome based on vote counting of the direction of the effect. Vote counting, a simple method for analyzing evidence from multiple evaluations, involves comparing the number of studies showing benefit (reduced MNA in the case of our study) with the number of studies showing harm (intact/increased MNA in the case of our study) (69). A harvest plot was designed to present results from the analysis. We also designed graphics to represent evaluated domains of methodological and reporting quality for each study and the quality across all studies.

To test for the statistical significance of the vote counting analyses, we used the sign test. The sign test is a non-parametric test that uses a binary measure of either a positive or a negative effect to test whether there is sufficient evidence to reject the null hypothesis of an equal number of positive and negative results (70). The P-value from a sign test represents the probability of observing the given number of positive and negative results if the null hypothesis was true. To perform the test, we counted the number of studies in each effect direction for the outcome. Also, to explore the results of the most commonly used paradigms, we conducted separate analyses on paradigms with more than 5 studies, which were EEG with 7 studies and task-based fMRI with 9 studies. We used GraphPad (Link) to calculate the two-tailed P-value for the sign test. We considered a p < 0.05 as significant (alpha error).

To explore heterogeneity in the results, we compared the outcome between subgroups. We conducted a test for subgroup differences between studies that evaluated MNA in “drug-naïve/drug-free for at least 1 month” patients, against studies on “medicated” patients. To check for this difference, we conducted Fisher's exact test (Link). Also, knowing that gray matter volumes atypically decline with age in schizophrenia patients (71), we conducted a logistic meta-regression test by comparing the mean age of the participants in each study, against the direction of the effect. We used the weighted least squares (WLS) method for this regression, with the weight associated with each study being the square root of its sample size (N). Similar subgroup analyses were done for the gender of the participants (female to male ratio), mean positive PANSS scores of patients, and mean negative PANSS scores of the patients against the direction of the effect.

To evaluate the robustness of our results, we conducted a sensitivity analysis by excluding studies that were judged to be of poor methodological and reporting quality. We used the same previous methods above for this analysis.

A summary of the characteristics of the included studies is presented in Table 1. The comments section for this table is built upon the comments provided in a similar table in the systematic review of Mehta et al. (54). For a more detailed report of the characteristics of contributing studies, see Data sheet 3.

MEG, fMRI, and PET are known to provide good spatial resolutions. The different patterns of activity in MNA-specific brain regions between cases and controls in the included studies are provided in Table 2.

IFG was the most investigated area across the literature where 6/7 studies reported the detection of a decreased MNA in that region. IPL was the second most investigated area, but interestingly, it was also the one with the most controversial results. Of the 6 studies that evaluated this area, 3 reported the detection of decreased MNA and the other 3 reported the detection of increased MNA. Also, MNA was reported to be decreased in PMv in 3 of the 4 studies that investigated this area. Results for the STG area were pretty consistent with 3 of 3 studies reporting decreased MNA. Only 2 studies reported a difference in the insula activation, where their results were in the opposite direction.

The results of the analysis for the primary outcome are presented as a harvest plot in Figure 4. Most studies concluded that MNA was significantly reduced in schizophrenia patients, compared to controls (23/29, 79.3%). The two-tailed sign test P-value was calculated to be 0.002, meaning that the chance of observing either 23 or more studies, or 6 or fewer studies in 29 studies, in that direction, is 0.2%. Only two studies (87, 92) found significant results in the opposite direction (2/29, 7.9%). Four studies (74, 75, 82–84, 102) concluded that there was no significant difference between patients and healthy controls (4/29, 13.8%).

We also conducted a vote-counting analysis for the direction of the effect for studies that only used task-based fMRI as their assessment tool. In this group, seven studies concluded that MNA was reduced in cases, although this finding was not statistically significant (7/10, 70.0%; P = 0.344). A similar analysis was also conducted for studies that only used EEG. In this group, four studies concluded that MNA was reduced in cases (4/7, 57.1%; P = 1.000), showing almost no statistical significance. Also, two studies showed an intact MNA and one concluded that MNA was increased in cases. Results were very contradictory for the EEG group and demanded explicit evaluation.

Regarding the primary outcome, for the patients in the “drug-naïve/free for at least 1 month” subgroup, 4/4 studies, and the patients in the “medicated” subgroup, 14/20 studies were in the direction of decreased MNA, while 6 studies were in the direction of either intact or increased MNA. The test for subgroup differences revealed no significant difference between them (P = 0.539).

Results for the logistic meta-regression analyses for age, gender (female to male ratio), positive PANSS scores, and negative PANSS scores against the direction of the effect for the primary outcome are presented in Table 3 and Figure 5. There seems to be a relationship between the age of participants and the direction of the effect. A similar relationship was observed between the positive PANSS scores and the direction of the effect. Studies that found intact/increased MNA, were performed on patients of higher age and higher positive PANSS scores. These relationships were found to be statistically significant (P < 0.001 for age and P = 0.004 for positive PANSS scores).

To check the robustness of our results for the primary outcome, we performed an analysis on studies that were judged to have fair or good methodological and reporting quality. Most of these studies were in favor of decreased MNA in cases, although this finding was not statistically significant (13/18; P = 0.096).

Mehta et al. (54) conducted a systematic review on the same subject in 2014. However, to our knowledge, this is the first systematic review that has also incorporated qualitative data analysis to evaluate the mirror mechanism in schizophrenia patients and identify some of the possible sources of heterogeneity in the findings.

Some hypothesize that schizophrenia might be a disorder of the “social brain” (108). Mirror neurons are collections of neurons that are believed to be part of this social brain network (109). From this point of view, it has been hypothesized that MNA is impaired in schizophrenia. Our study reveals, with moderate confidence, that there is indeed an impaired MNA system in these patients.

Although most findings were in the same direction, one might question why others found contradictory results. We aim to describe here some of the potential causes underlying those results.

First, our meta-regression analyses found a statistically significant relationship between the mean positive PANSS scores and age of the participants in each study and the direction of the observed effect in that study. Studies that demonstrated intact/increased MNA enrolled patients with higher positive PANSS scores and age, compared to the studies that demonstrated decreased MNA.

Regarding the relationship between the positive PANSS scores and the direction of the effect, our results indicate that patients with higher positive PANSS scores are more likely to have intact/increased MNA. Positive PANSS measures the severity of the positive symptoms of the disorder, such as delusions, conceptual disorganization, hallucinations, and hostility (62). McCormick et al. (87) found a similar pattern in their study, suggesting that MNS may be overactive when positive symptoms are most prevalent (especially hallucinations). Mitra et al. (89) reported a negative correlation between the mu wave suppression and the thought disturbance cluster on PANSS, proposing that according to the theory that dopamine levels in the brain and the performance of the brain circuits have an “inverted-U” shaped relationship, an increase in brain dopamine levels during schizophrenia possibly disrupts the MNS circuit, leading to psychopathology manifestations. Other studies did not find a significant correlation between positive PANSS scores and MNA (74, 78, 80, 84, 86, 88, 90, 96, 99, 102). These contradictory results could be due to differences in experimental conditions, stage of disease, or the measures used to assess symptoms. Nevertheless, the idea of MNA correlating with patients' symptoms seems to be a plausible hypothesis. Indeed, this hypothesis was previously mentioned by Mehta et al. (54), making it an explanation worth further investigation. This is also in line with Frith and Corcoran's theoretical model of the relationship between social cognitive processes and psychotic symptoms (110).

Regarding age, similar results were found previously in autism, suggesting that individuals with autism may outgrow any mirror neuron deficit after a certain age (111, 112), although some other studies question these results (113, 114). One recent study indicates that in general, there might be some differences in the MNA between younger and older adults (115), where older adults showed Mu suppression in frontal and frontotemporal regions during a memory task, in contrast with young adults who showed Mu enhancement. Besides these, some studies have also shown that the social cognitive performance of schizophrenia patients may actually increase by age (116). Linke et al. (117) found a similar pattern in their study as well, but after including the patient's age at onset in their models, they concluded that this observed increase in social cognitive performance is not really due to the patients' age, but it is actually due to their later onset of the disease, as older patients are usually those with a later onset of psychosis as well. This is in line with previous studies that revealed age at the onset of the disease is negatively correlated with patients' cognitive performance (118, 119).

In the study of Horan et al. (82–84), they used a mask before group-level analyses, which might “bias against finding significant between-group differences,” as stated by the authors. In the study of Andrews et al. (73), they used a combination of TMS and EEG that may have reduced the quality of the EEG signals from some participants. Also, the baseline stimulus used to directly compare the EEG and TMS measures (blank screen) was not the same for the two measures. The studies of McCormick et al. (87) and Quintana (92) found increased MNA in patients. Quintana (91) study made a controversial decision by excluding BOLD signal changes during incorrect responses. The authors proposed that patients may have a compensatory increase in MNA while correctly performing the task. In the study of McCormick et al. (87), subgroup analyses showed only a subgroup of patients had greater mu suppression, the active psychosis subgroup. These findings indicate the need for more research on these subgroups of schizophrenia patients.

Most notably, we found the results of EEG studies to be very contradictory. Some possible explanations for such results have been previously mentioned in the study of Hobson and Bishop (120). First, they suggested that because the mu frequency band overlaps with the alpha frequency band (which is sensitive to attentional fluctuation), mu suppression could potentially be confounded by changes in attentional engagement. They also report that there is little consistency in how the specific baseline against which mu suppression is assessed should be defined. Finally, they examined mu suppression in 61 typical adults and reported that even in an optimal evaluation condition, 16–21% of participants showed no mu suppression to action observation task. Overall, they concluded that mu suppression can be used to index the human MNS, but the effect seems to be weak and unreliable, and it may also easily be confounded by alpha rhythm suppression. More interestingly, a recent study found that observation tasks may sometimes elicit mu rhythm enhancement rather than suppression (121). All these results question the reliability of the EEG paradigm for assessing MNA. Also, the validity of the TMS/EEG paradigm has been seriously questioned by another recent study (122). With all of those in mind, we still didn't consider these paradigms as invalid in our bias assessment process, as this domain required subjective judgments (where we tried to be conservative) and there are still some counter-arguments supporting the possible reliability of these paradigms.

Negative symptoms account for a substantial portion of the morbidity associated with schizophrenia (123). Empirical research has argued for an association between negative symptoms and anomalous MNA (124). We found the same association in some of the studies included in this review (85, 86, 95, 98). The study of Singh et al. (95) found lower mu wave suppression to positively correlate with negative PANSS scores, suggesting MNS may be underactive when negative symptoms predominate. However, the study of Brown et al. (76) found a statistically significant correlation between mu wave suppression and negative PANSS scores in the opposite direction of Singh's et al. Also, the study of Kato et al. (85) reported a negative correlation between the amplitudes of root-mean-square (RMS) of MEG responses and negative PANSS scores. Finally, Park et al. (91) reported the presence of a negative correlation between the functional deficits in MNS and negative PANSS scores. Although we didn't find any significant correlation between MNA and negative PANSS scores (P = 0.226), future studies should provide an in-depth assessment of the relationship between these two factors.

Deficits in communications skills have been previously documented in schizophrenia (125), but there has not been a comprehensive explanation for the etiology of this phenomenon up to this date. Indeed, MNS has been linked to developing communication skills via integrating auditory, visual, and motor stimulation (126). A study by Cantisani et al. reported a negative linear association between resting-state cerebral blood flow in the left inferior and middle frontal gyrus of schizophrenia patients with their communication skills, measured through the Social and Occupational Functioning Assessment Scale (SOFAS) (127). Our results indicate that across the literature, the inferior frontal gyrus (IFG) was the most investigated area for MNA in schizophrenia patients, where most studies indicated decreased MNA detection in this area. Putting these findings together, the disruption in MNA might be suggested as a possible explanation for communication skills in schizophrenia patients. Further studies are required to validate this hypothesis.

Echopraxia is the pathological repetition by imitation of the movements of another person. In the context of schizophrenia, it has been mostly associated with the catatonic form (128). A previous speculative paper by Pridmore et al. suggested that pathologically handled MNS-generated representations, especially in IFG, might be involved in this dysfunction (129). Indeed, this was in line with the findings of the study of Zaytseva et al. (101) where the authors reported altered mu rhythm suppression in the right frontal and central brain regions in patients with catatonic schizophrenia. More studies on catatonic patients in the future are suggested to further evaluate the validity of this finding.

Previously, some have argued that MNS might have a plastic feature (130), meaning that after receiving treatment, disruptions in this system might at least partially resolve. However, a study by Mitra et al. (131) found that following 8 weeks of antipsychotic treatment, no significant changes took place in the MNA of patients. This is partly in line with our results that revealed there was no significant MNA difference between medicated and drug-free patients. These indicate that even though antipsychotic medications may improve cognitive deficits for some schizophrenia patients, they may not affect MNA significantly.

Our results indicate that MNA is altered in schizophrenia patients, similar to the individuals with autism. This finding contributes to the efforts of exploring the dimensions of mental disorders to integrate many levels of information to understand the nature of mental health and illness, such as efforts taking place in the projects of RdoC (132).

Deriving clinical impact from such results could be an existing area of research. In a pilot study in 2020 (133), Hadoush et al. evaluated the effect of bilateral anodal transcranial direct current stimulation (tDCS) applied over the MNS of autism patients. They concluded that this intervention has a moderate therapeutic effect on children with autism in terms of their sociability, behavior, health, and even physical conditions. This pilot study reveals the potential of new rehabilitation methods through MNS-based training, which might benefit patients. It might be interesting to evaluate if similar results could be obtained for schizophrenia patients.

Another study that evaluated the effect of add-on yoga therapy on schizophrenia patients, revealed that MNA increased in the intervention group following 6 weeks of yoga therapy. They also found significant improvements in social cognition composite score (SCCS), negative symptoms (SANS), and positive symptoms (SAPS). One hypothesis is that the improvements in those clinical symptoms might have been achieved through the training of the MNS. Indeed, a previous study on yoga therapy for 2 years on 12 autism patients (6 in the interventional arm and 6 in the comparator arm) revealed improvements in imitation and other social skills of the participants (134). The authors hypothesized that guided imitation of therapist body positions might have stimulated MNA, resulting in an improved sense of self. Investigating the causal relationship between such findings might benefit future research.