Shared and distinct functional networks for empathy and pain processing: a systematic review and meta-analysis of fMRI studies

Abstract Background Empathy for pain is a complex phenomenon incorporating sensory, cognitive and affective processes. Functional neuroimaging studies indicate a rich network of brain activations for empathic processing. However, previous research focused on core activations in bilateral anterior insula (AI) and anterior cingulate/anterior midcingulate cortex (ACC/aMCC) which are also typically present during nociceptive (pain) processing. Theoretical understanding of empathy would benefit from empirical investigation of shared and contrasting brain activations for empathic and nociceptive processing. Method Thirty-nine empathy for observed pain studies (1112 participants; 527 foci) were selected by systematic review. Coordinate based meta-analysis (activation likelihood estimation) was performed and novel contrast analyses compared neurobiological processing of empathy with a comprehensive meta-analysis of 180 studies of nociceptive processing (Tanasescu et al., 2016). Results Conjunction analysis indicated overlapping activations for empathy and nociception in AI, aMCC, somatosensory and inferior frontal regions. Contrast analysis revealed increased likelihood of activation for empathy, relative to nociception, in bilateral supramarginal, inferior frontal and occipitotemporal regions. Nociception preferentially activated bilateral posterior insula, somatosensory cortex and aMCC. Conclusion Our findings support the likelihood of shared and distinct neural networks for empathic, relative to nociceptive, processing. This offers succinct empirical support for recent tiered or modular theoretical accounts of empathy.


Introduction
Empathy is a critical concept in human emotional and social experience. The ability to share in the affective states of those around us brings evolutionary advantages, enabling us to respond to the needs of others, predict their behaviour and support decision-making about our own actions (Decety et al., 2012). Empathy for observed pain leads to the generation of mental health and well-being and impacts on societal structures (Bernhardt and Singer, 2012).
Nociceptive (pain) processing is associated with widereaching patterns of neural activation which briefly encompass bilateral anterior, mid-and-posterior insula cortices, primary and secondary somatosensory cortex, inferior frontal gyri (IFG) and supramarginal gyri, as well as medial clusters in anterior cingulate/anterior midcingulate cortices (ACC/aMCC), thalami and brainstem (Melzack, 2001;Wager et al., 2013). Previous neuroimaging studies highlighted a similar pattern of neural activation during empathy for pain, with particular overlap between direct experience of pain and empathy for pain described in ACC/aMCC and anterior insula (AI) cortex (Morrison et al., 2004;Singer et al., 2004;Jackson et al., 2005). As such, it was posited that shared representations in the brain for empathy and direct pain experience could underpin the similarities in neural activation profiles for both (Preston and de Waal, 2002;de Vignemont and Singer, 2006).
The Perception-Action Model (PAM) of empathy (Preston and de Waal, 2002), suggests that attended perception of another's state automatically activates one's own subjective representations, leading to generation of associated responses unless inhibited. As well as aforementioned neuroimaging studies, research in support of the PAM highlighted the automaticity of empathic responses from humans  through to lower order mammals (Langford et al., 2006). Such automaticity may reflect an inherent dependence on specific neurobiological mechanisms e.g. mirror neuron systems (Rizzolatti and Craighero, 2005) which could facilitate the close relationship between perception and action responses.
However, current definitions of empathy suggest the involvement of automatic affective processing but also include aspects of higher order cognition. For example, empathy requires isomorphic sharing of feelings with another, but also necessitates awareness that one's state originates from observation of the target (Singer and Lamm, 2009). In support of this, neuroimaging research has shown that empathic brain responses in humans are modulated by top-down psychological factors including (but not limited to) situational context (Singer, 2006), intentionality (Akitsuki and Decety, 2009), in/out-group status (Azevedo et al., 2014), implicit bias (Azevedo et al., 2013), pain catastrophising (Fallon et al., 2015) and relevant expertise (Cheng et al., 2007). Topdown modulation of empathic brain responses points to processes beyond mere shared representation, which would likely necessitate a pattern of brain activation that is at least partly distinct from nociceptive pain processing.
The Russian-Doll model of empathy posits a tiered system with progressive levels of empathy from basic affective (e.g. emotional contagion) to higher order processes such as sympathetic concern and emotional perspective taking (de Waal, 2008). Similarly, other proponents also suggest multifaceted neuroscientific theoretical models of empathy which necessitate shared representations between self and other, plus the ability to distinguish between the two, as distinct and essential building blocks for empathic experience . Likewise, the most recent iteration of the PAM describes a dynamic and graded system with flexibility for learning and experience which is likely to recruit different brain regions for distinct components of empathic processing (de Waal and Preston, 2017). Novel approaches are required to provide empirical data to support understanding of brain regions or networks of regions involved in shared and distinct theorized components for empathy and direct pain processing.
Neuroimaging studies demonstrate a heterogeneous profile of activation foci for empathy for pain (Zaki et al., 2016), which may be due in part to fundamental issues with sample size, study designs and the mass univariate approach typically employed in fMRI studies . Meta-analysis represents a pertinent tool for understanding the social neuroscience of empathy due to its inherent advantages for overcoming issues including design heterogeneity and small sample sizes (Button et al., 2013). Earlier meta-analyses of empathy for pain highlighted a conjunction for direct pain experience and empathy for pain in a core network of brain ACC/aMCC and AI bilaterally (Lamm et al., 2011). Similar regions were also seen in a coordinate based meta-analysis of empathic processing including, but not restricted to, pain (Fan et al., 2011) and empathic processing for empathy which excluded empathy for pain modality .
More recent coordinate based meta-analyses have confirmed core activations for empathy for pain in AI and ACC/aMCC. However, they also revealed additional activations profiles in postcentral gyrus, inferior parietal lobe and deep brain structures including thalamus and brainstem (Timmers et al., 2018) bilateral IFG, supramarginal and fusiform regions and the right anterior lobe of the cerebellum (Jauniaux et al., 2019;Xiong et al., 2019). Given the likelihood of an extended network of empathic processing, it would now be advantageous to consider spatial conjunction and contrast for empathy for pain and processing of actual pain. Although neuroimaging studies have previously considered conjunction (Lamm et al., 2011), until now, there is no meta-analysis of brain imaging research to investigate the spatial contrast between empathy for pain and direct pain experience.
Furthermore, the current interpretation of existing data, which points to overlap of empathy and direct processing for pain in aMCC and AI, is further complicated by the relevance of these regions for broader states that are independent of pain processing such as interoception, arousal and attention (Craig, 2009). Indeed, these regions are statistically the most commonly activated across an expansive range of fMRI experiments with varying themes (Yarkoni et al., 2011). Moreover, the specificity of pain-related activations in AI and aMCC was recently criticized (Iannetti et al., 2013), which in turn casts doubt on the relevance of conjunction activations with empathy for pain. Taken together, we need to consider new empirical approaches which will allow for an improved understanding of how activation profiles for empathy for pain relate to, and differ from, those associated with direct experience of pain to inform our theoretical understanding of empathic processing.
To achieve this aim, the present study proposed to compare brain activations associated with empathy for observed pain and nociceptive processing by employing a coordinate based metaanalysis with activation likelihood estimation (ALE, Eickhoff et al., 2009;Eickhoff et al., 2012). A principle aim of our metaanalysis was to consider the whole-brain network of brain activations associated with empathy for observed brain whilst minimizing bias, we therefore opted to exclude studies which only utilized region of interest (ROI) analysis which may bias meta-analyses towards more established or accepted regions Turkeltaub et al., 2012). In addition, we expanded on previous investigations by performing conjunction and contrast analyses of our ALE findings with those of a recent comprehensive coordinate based meta-analysis of pain processing (Tanasescu et al., 2016). This allows, for the first time, an empirical meta-analysis evaluation of shared, and distinct, activation profiles for empathy for pain and direct pain experience.
Following-on from recent extended meta-analyses, we hypothesized that empathy for observed pain would demonstrate a greater degree of overlap with nociceptive pain processing which would extend beyond established core activations in AI and aMCC. In addition, we posited that empathy for pain and direct experience of pain would each demonstrate distinct activation profiles in line with recent theoretical descriptions which highlight the likelihood of modular components of empathy with potential for hierarchical functional relevance de Waal and Preston, 2017). We anticipated that empathy for pain, relative to actual pain experience, would demonstrate increased likelihood of activation in brain regions associated with higher order processing and particularly selfother distinction such as parietal cortices and temporoparietal junction (TPJ). Conversely, it was expected that direct pain experience would elicit greater likelihood of activations in brain regions such as posterior insula which receives afferent nociceptive projections (Garcia-Larrea, 2012) and therefore would not be recruited by empathy for pain.

Data search and extraction
The formal search strategy was conducted according to the PRISMA guidelines for reporting meta-analyses and systematic reviews (Moher et al., 2009). This consisted of systematically searching four electronic databases during October 2019 (Medline, Pubmed, PsycINFO, Scopus) using the MeSH search terms (magnetic resonance imaging OR fMRI) AND (functional OR brain activation OR neural activity OR BOLD) AND (Pain) AND (empathy OR empathic). Searches were restricted to terms found in the title or abstract of articles. No date limit was set for the searches and manual searches of the reference sections of identified papers were conducted to supplement the formal search process. Previous meta-analyses of empathic processing (Fan et al., 2011;Lamm et al., 2011;Timmers et al., 2018;Jauniaux et al., 2019) were also screened for additional articles although this did not lead to the inclusion of any further studies.
Article selection and extraction of data. Formal database searches were conducted by two authors independently (N.F. and C.R.), as were supplementary and manual searches. Both authors were responsible for assessment of articles for inclusion, and decisions over article inclusion were determined by discussion, disagreements where resolved via discussion or presented to a third arbiter (A.S.). One author (N.F.) extracted the relevant coordinate data, which was cross-checked and confirmed by a second (C.R.). Studies that reported coordinates in the Talairach space were converted into MNI using GingerALE software for the purposes of analysis and reporting.
Eligibility criteria. The criteria for inclusion were: (i) human fMRI studies published up until October 2019; (ii) original English language articles; (iii) published in a peer-reviewed journal; (iv) utilizing a paradigm including visual pain stimuli i.e. images, videos or animations of pain scenes or pain facial expressions; (v) employed an appropriate contrast with a suitable control stimulus (e.g. non-painful scene, animation, etc.); (vi) coordinates were reported in the paper or supplementary material of the direct (pain > non-pain) contrast in either Montreal Neurological Institute (MNI, (Evans et al., 1993)) or Talairach space, studies that reported coordinates in the Talairach space were converted into MNI for the purposes of analysis and reporting; (vii) data were obtained from a healthy population without experience or expertise that could modulate empathic processing (e.g. we excluded studies that explicitly recruited only clinicians or other healthcare professionals or only incarcerated individuals); (viii) whole-brain analysis were reported with thresholding of (or equivalent to) P < 0.001 uncorrected voxelwise throughout the whole brain with at least P < 0.05 cluster level correction (or equivalent) declared. We excluded papers which only reported ROI results.
Meta-analysis has shown that the selected level of hierarchical thresholding is optimal for balancing sensitivity to effect with reduced risk of false positives (Woo et al., 2014). This level was selected to reduce bias by reducing the inclusion of false positive reports which are highly likely to favour specific regions that are established in the literature. This was also the reason for excluding studies which only utilized ROI analyses. In few potentially acceptable studies when no thresholding was reported, we contacted individual groups to request details but received no replies. For comparison with direct pain experience, we utilized an existing open-access database of coordinates included in a previous meta-analysis of experimental pain stimulation during fMRI (Tanasescu et al., 2016). Inclusion criteria were similar to those in the present study. Briefly, the authors only included research that reported whole-brain group analysis of a pain induction>baseline contrast. From the full dataset (which included subgroups of chronic pain populations and also healthy people with chemically induced hyperalgesia) we extracted only studies which focused on cutaneous pain in healthy populations to reflect the population extracted for the present empathy meta-analysis. Further details on the studies included can be found in Supplementary Materials 3.
In order to determine consistency in reported regions of activation for both analyses, ALE meta-analyses was performed in Brainmap GingerALE v2.3.6 (Eickhoff et al., 2009). This method assigns an ALE value to each voxel throughout the whole brain, with greater ALE values indicative of more studies which report activated peaks at a voxel or in close proximity. We implemented the more stringent correction of ALE estimation (Turkeltaub et al., 2012). This method utilizes a random effects model to minimize within-experiment effects (by accounting for the number and proximity of reported foci) and within-group effects (by accounting for multiple contrasts reported from a single study), both of which can unduly influence meta-analyses. This method therefore optimizes the degree to which ALE values represent concordance of findings across independent contributions.
Standardized procedures and default parameters were employed throughout the analysis adhering to recent guidelines on methodology for appropriate study selection and thresholding of results (Eickhoff et al., 2016). The concordance of ALE values throughout the brain for empathy for pain (empathy > control), and direct processing of pain (pain > non-pain) were evaluated in comparison to random distributions using permutation analysis (Maris and Oostenveld, 2007) with 10 000 permutations. Deactivation contrasts (e.g. control > empathy, non-pain > pain) were not included due to the infrequency of reporting of this direction of contrast in the literature. An initial cluster forming threshold (uncorrected P < 0.001) was implemented followed by cluster-level Family-wise error (FWE) correction (P < 0.05) to identify relevant ALE regions as recommended in recent publications (Eickhoff et al., 2016). Secondly, resulting ALE maps for empathy for pain and direct pain experience were compared using conjunction and contrast analyses. Again permutation analysis was performed with 10 000 permutations to identify brain regions which demonstrated overlapping and distinct patterns in the respective ALE maps for each process. For cluster analysis, results were thresholded using cluster-level false discovery rate (FDR, P < 0.05) and a minimum cluster size of 200 mm 3 as previously recommended (Eickhoff et al., 2016). FDR correction is the most stringent correction method available in Ginger ALE software for contrast analyses, FWE method is not available for contrast analyses (Hoffman and Morcom, 2018;Papitto et al., 2020). Figure 1 illustrates a flowchart indicating the study selection steps, (see Supplementary Material for spreadsheet of search process and PRISMA checklist). Specifically, a total of 502 articles were returned from the initial searches (Medline; 124; Pubmed; 154, PsycInfo; 129, Scopus; 95). Of these, 304 were duplicates from repeated searches and removed in the first step. A further 75 articles were removed following the initial review of titles and abstracts. Studies excluded at this stage included: those where it was clear and obvious that no suitable (i.e. healthy adult) population was reported (16), studies where it was clear and obvious that they did not utilize observed pain stimuli (30), not an experimental report (e.g. review articles) (22), not fMRI method (7). Furthermore, following full-text review a further 84 articles were removed including those which exhibited an inappropriate contrast (e.g. comparisons that did not match the selection criteria, in-group empathy > out-group empathy) or never reported the basic contrast of interest (empathy > control) (37), those which never met the thresholding criteria (13), did not utilize observed pain stimuli (16), never reported a healthy, adult and non-expert population (11), were not an experimental report journal article (2) or which only reported ROI analyses (5), leaving a total of 39 studies for the analyses of empathy for pain (Table 1).

Significant ALE clusters for empathy for pain.
The empathy for observed pain-non-pain contrast ALE metaanalysis pooled data from a total of 1112 participants and 527 reported foci. The results (Table 2) revealed 10 significant clusters where ALE values demonstrated significant levels of consistent spatial activation for empathy for observed pain. Two large bilateral clusters covered AI extending laterally to IFG with a smaller independent cluster also evident in right IFG. Bilateral clusters were identified in supramarginal and lateral occipitotemporal cortices and a medial cluster was elicited in ACC/aMCC extending superiorly to supplementary motor cortex. Figure 2 illustrates the location of significant ALE clusters from the meta-analysis of empathy for pain.

Directly perceived pain
The findings of the coordinate based analysis for directly perceived pain studies reflect the results from the existing research where they are described, illustrated and discussed in more detail (Tanasescu et al., 2016). To briefly summarize, our ALE meta-analysis contrast of experimental pain in healthy participants-baseline pooled data from 180 eligible studies with a total of 2605 participants and 2780 reported foci. Results revealed four (extensive) clusters of activation encompassing primary and  secondary somatosensory cortices (S1, S2), bilateral anterior, mid and posterior insula, prefrontal and premotor cortices, bilateral putamen and thalami and medial clusters in ACC/aMCC and brainstem regions.

Conjunction analysis
The conjunction analysis of ALE maps representing empathy for pain and direct pain experience pooled data from 219 studies with a total of 3717 participants and 3307 reported foci. The results highlighted an overlap of activation likelihood coordinates in seven clusters encompassing bilateral AI and ACC/aMCC as well as bilateral IFG which bordered (and in the case of the right hemisphere extended to) middle frontal gyrus, and bilateral supramarginal regions (Table 3, Figure 3).

Contrast analyses: empathy-pain
Contrast analysis comparing the ALE maps of concordant activations for each process pointed to significantly greater likelihood of activation during empathy for pain relative to directly perceived pain in 6 clusters encompassing bilateral supramarginal, IFG and occipitotemporal regions (Table 4, Figure 4).

Contrast analyses: pain-empathy
The reverse contrast revealed six clusters indicative of increased activation likelihood estimates for directly perceived pain relative to empathy for observed pain. These regions encompassed two large bilateral clusters encompassing parietal opercular cortices (S2), posterior insula and S1. Right putamen was also evident for directly perceived pain relative to empathy. A right frontal cluster encompassing right prefrontal and dorsolateral prefrontal cortices was also elicited. Two medial clusters demonstrated increased concordance of activation in direct, relative observed, pain in aMCC (Table 5, Figure 4).

Discussion
The findings of the ALE meta-analysis of empathy for pain revealed concordant activations to observed pain stimuli located in aMCC and bilateral AI which accords with previous investigations (Fan et al., 2011;Lamm et al., 2011). However, the present results expand on previous analyses to reveal a richer and more complex pattern of activation encompassing bilateral supramarginal, IFG and occipitotemporal cortices. It is noteworthy that previous research did allude to the potential for a broader range of contributing brain regions including some of these specific regions (Lamm et al., 2011). Moreover, such a restricted profile of activation is inconsistent with the subjective complexity of empathic experience and the heterogeneity of activation foci seen across relevant fMRI studies (Zaki et al., 2016). Therefore, the present analysis aligns with recent meta-analyses to confirm the existence of previously established core activations in AI and aMCC (Fan et al., 2011;Lamm et al., 2011;Bzdok et al., 2012) for empathic processing, but also indicates the likelihood of additional activations in bilateral inferior parietal lobe, IFG and occipitotemporal cortices (Timmers et al., 2018;Jauniaux et al.,  2019; Xiong et al., 2019). Taken together, this body of research goes some way towards establishing neurobiological underpinnings of empathy by highlighting the involvement of an extended network beyond these core brain regions. In the present analysis, all brain regions described in results were significant contributors to the ALE of empathy for pain after implementing stringent and recently recommended guidance regarding statistical thresholding (Turkeltaub et al., 2012;Eickhoff et al., 2016;Eickhoff et al., 2017) and study selection (Moher et al., 2009;Woo et al., 2014). Moreover, left supramarginal gyrus was amongst the most frequently observed regions in terms of contributing studies, even surpassing right AI, which suggests that the accepted predominance of core regions does not truly reflect the neurobiological underpinning of empathy for observed pain. The present study was the first to utilize coordinate based meta-analytical methods to statistically evaluate the conjunction and contrast between empathy for observed pain and direct painful experience. For this aim, we compare ALE maps of empathy for pain with those of a conclusive analysis of nociceptive processing comprising some 180 experimental pain studies in healthy people (Tanasescu et al., 2016). Similar to the extended pattern of foci observed for empathy for pain, conjunction analysis highlighted a broad range of overlap for empathy for pain and direct experience of pain with bilateral clusters in AI, IFG and supramarginal gyri, as well as medial clusters in ACC/aMCC. These regions are all important for pain processing, and highlighted in the recent fMRI neural signature of pain (Wager et al., 2013) and its predecessor, the pain neuromatrix (Melzack, 2001). The conjunction between pain and empathic processing builds upon previous image-based analyses which suggested that spatial overlap was primarily limited to AI and aMCC, although this previous study exclusively utilized cue-based, as opposed to visual observation of pain, paradigms (Lamm et al., 2011). The present results go further and support proponents of the PAM (Preston and de Waal, 2002;de Waal, 2008;de Waal and Preston, 2017) by demonstrating a broader functional network of brain regions which occupy the same morphometric space and are recruited during observed pain phenomena and direct experience of pain. This provides empirical support for theoretical interpretations which recognize a degree of shared representation between empathy and direct experience of pain (Singer and Lamm, 2009;Bernhardt and Singer, 2012;de Waal and Preston, 2017;Lamm et al., 2017). In summary, we believe that the present meta-analysis represents the most succinct evidence to date of an extensive conjunction for pain experience and empathy for pain, encompassing a diverse array of pain relevant brain regions in a bilateral pattern indicative of a coherent network. However, it is important to note that evidence of conjunction of processing between nociceptive and empathic pain does not necessarily imply the same psychological representation (Zaki et al., 2016). For example, debate around functional specificity and complexity for pain (and associated) processing in anterior cingulate region is well documented (Lieberman and Eisenberger, 2015;Wager et al., 2016). Contrast ALE analyses to compare concordant activations that exhibited greater likelihood for empathy for pain compared to direct experience of pain (and vice versa) were also performed for the first time. Empathy, compared to direct pain experience, demonstrated preferential bilateral activation in supramarginal regions, which extended superiorly to the supramarginal gyrus. Although supramarginal activations are frequently reported in fMRI empathy literature, their specific relevance is often not subject to discussion or interpretation, particularly if the basic empathy contrast is not the primary aim of the research (Jackson et al., 2006;Guo et al., 2013;Fan et al., 2014;Berlingeri et al., 2016;Zheng et al., 2016;Benuzzi et al., 2018). Of the many studies contributing supramarginal activations to the present ALE analysis, few discussed their relevance. Those which did interpreted the activation in terms of understanding and anticipating pain (Azevedo et al., 2013), or appraising unpleasantness . Others designated this activation as S2, and proposed a more sensory role (Ma et al., 2011). The consistency of bilateral activation of supramarginal regions for empathy suggests a need for greater understanding of its role. fMRI research has demonstrated that supramarginal activity can represent recognition of noxious environmental stimuli (Benuzzi et al., 2008) and reorientation of attention to threat Carter and Huettel, 2013). This potentially offers a mechanistic explanation of activity in supramarginal gyrus during empathy for observed pain stimuli, which could underpin the previously proposed higher order interpretations regarding anticipation or valence appraisal Azevedo et al., 2013) which would be required for identification and orientation to threat.
Contrast analyses also pointed to bilateral occiptotemporal activations that were more likely to be engaged during observed pain, but not direct pain experience. Previous fMRI research have explained these activations for empathic viewing in terms of enhanced visual processing (Azevedo et al., 2013) likely due to augmented salience (Akitsuki and Decety, 2009) or attention (Han et al., 2017) for pain scenes. As with supramarginal activation patterns, such lateral occipitotemporal activations are often reported with minimal or no discussion (Vachon-Presseau et al., 2012;Morelli et al., 2014;Seara-Cardoso et al., 2015;Enzi et al., 2016;Fourie et al., 2017;Richins et al., 2019), despite often being amongst the strongest research results in terms of statistical significance. Lateral occipitotemporal cortex encompasses the extrastriate body (EBA) which is associated with perception of body parts (Downing et al., 2001). As the present meta-analysis which focused on paradigms which included visual depictions of pain scenes, most often to limbs or face, this seems reasonable. However, all studies included used a suitable visual control, i.e. same body part in absence of pain, which suggests some empathy (or at least salience-related) function underlying these augmented bilateral activations during empathic processing. Interestingly, activation in lateral occipitotemporal cortex, and particularly EBA, were previously shown to reflect in-group status during empathy for pain (Azevedo et al., 2013) although another study did not report empathy activation in EBA (Lamm and Decety, 2008). Recent research has demonstrated relevance of EBA for self-other recognition of body parts (De Bellis et al., 2017), which suggests potential for the region in higher order aspects of empathic processing, particularly when observing visual depictions of others undergoing somatic pain. The present findings suggest that the occurrence of activation in these regions is frequent, but not assured. Occipitotemporal activations were identified in 30-40% of fMRI research with no clear pattern suggesting that the activation relates to a specific paradigm or stimuli type. Therefore, the nature of lateral occipitotemporal processing during empathy for pain warrants further investigation. Contrast analyses also revealed greater bilateral activation in ventral IFG during empathy for pain relative to direct experience of pain. The IFG is frequently activated during motor imagery or action observation type paradigms (Hétu et al., 2013). As IFG is a core component of the mirror neuron system (Rizzolatti and Craighero, 2005), it has been easy to dismiss this as automatic, perception-action processing and a neighbouring part of this region was also highlighted by conjunction analysis. However, the contrast analysis significance of bilateral IFG in the empathy, relative to pain, contrast analysis suggests the likely importance for empathic processing. Previous studies demonstrated increased activation in IFG when participants were rating observed pain rather than a distractor task leading to the suggestion IFG could also contribute to higher order cognitive processes such as attaching meaning to the empathic situation (Gu and Han, 2007;Budell et al., 2010). Moreover, a recent empathy viewing study identified this region as relevant for the process of mentalizing similarity between oneself and the target that one is observing (Majdandzic et al., 2016). Therefore, our findings suggest that processing of visually observed somatic pain in another could recruit activation of IFG to facilitate elements of action-understanding and self-other processing. ALE of directly perceived pain, compared to empathy, demonstrated preferential concordance of activation in bilateral S1, posterior insula and parietal operculum, right putamen, right prefrontal cortices and aMCC. Posterior insula and parietal opercular cortices represent the primary targets of nociceptors in the spinothalamic tract (Garcia-Larrea, 2012) and the medial regions highlighted in the contrast reflect functional imaging of somatic aspects of pain perception such as gauging pain intensity (Coghill et al., 1999). Overall the contrast points to an absence of empathic processing in medial pain-processing regions, suggesting a pattern wherein empathy for pain shares more overlap with cognitive-affective aspects than somatic processing, as was previously hypothesized (Singer et al., 2004), albeit in a broader neural network than previously thought.
From a theoretical perspective, the patterns of ALE seen in conjunction and contrast analyses show alignment with a tiered theoretical understanding of empathic processing such as the Russian-Doll model (de Waal, 2008;de Waal and Preston, 2017) or independent components of shared representation and self-other distinction . The conjunction of activations in bilateral pain processing brain regions suggests a degree of automatic state-matching or emotional contagion  4. The location of significant clusters from contrast analysis of ALE maps for greater likelihood of empathy for pain relative to directly perceived pain and vice versa. Results in green indicate regions which showed greater likelihood of concordance of activation for empathy for pain, but not for direct pain experience. Results in blue indicate regions which showed greater likelihood of concordance of activation for direct pain experience, but not for empathy for pain. All clusters are overlaid onto standardized MNI template anatomical brain in (A). 3D surface projection. (B). As a montage of coronal slices throughout the whole brain. Relative Z scores are indicated by colourbar. (Preston and de Waal, 2002;de Waal, 2008). This is extended by empathy distinct activity in supramarginal, occipitotemporal and IFG regions which may reflect higher order aspects of empathy, and particularly mechanisms of self-other distinction. Previously, TPJ was highlighted for distinction of self-other during empathic processing (de Waal and Preston, 2017). Although our analysis did not indicate TPJ activations during empathy, this region shows greatest functional and anatomical connectivity with supramarginal gyrus and occiptotemporal cortices (Carter and Huettel, 2013;Igelström and Graziano, 2017). Previously, it was posited that TPJ activation could be associated with nuances of the empathic situation, particularly when it requires cognitive perspective-taking (de Waal and Preston, 2017). The present meta-analysis utilized paradigms eliciting empathy via observation of pain, which are less dependent on cognitive perspective-taking. Therefore, supramarginal and lateral occipitotemporal activations, which border TPJ, may reflect more automatic aspects of self-other distinction.
Possibly reflecting mechanistic processes relevant to self-other distinction including reorienting attention and quantifying threat Carter and Huettel, 2013) or interpreting self-other distinction of perceived body parts (De Bellis et al., 2017). We can speculate that the network of regions highlighted by the present, meta-analysis of empathy for observed pain is more relevant to these mechanistic aspects of empathic processing, whereas higher order cognition, such as emotional perspective taking, may recruit TPJ. The role of inferior parietal and occipitotemporal cortices for a broad range of social processing is a topic of considerable research, and a nexus of social processing extending from the angular gyrus of the TPJ anteriorly to supramarginal gyrus and posteriorly to occipitotemporal cortices was posited (Carter and Huettel, 2013). Others have highlighted the spatial relevance of TPJ and surrounding regions including supramarginal gyrus for a broad range of social cognition and self-other distinction (Igelström and Graziano, 2017). In light of the present metaanalysis, the pattern of activation in these parietal-occipital regions extending beyond the TPJ, which are often neglected in existing functional imaging research, could actually have important relevance for empathy for pain.
The present study has some limitations. As mentioned, we focused on empathy for observed pain rather than more complex iterations of cognitive empathy such as paradigms which utilize learning to associate abstract cues with pain stimulation delivered to another person located outside of the scanner. The former design is more prevalent in fMRI research (Lamm and Majdandžić, 2015), allowing for a greater degree of studies to be included. To include cognitive empathy paradigms would necessitate the permission of a smaller cohort of studies, falling below the minimum 17 independent studies recommended for ALE analysis (Eickhoff et al., 2016). Else it would mean permitting less stringent search and analysis methods which would sacrifice the integrity of the aims which focused on concordance across studies which utilized whole-brain analyses and robust statistical thresholding to reveal empathy for pain activations regardless of existing bias. However, it should be noted that previous meta-analyses suggest distinct patterns of activation for observed pain compared to more cognitive methods of evoking empathic response (Lamm et al., 2011) and specifically that observing visual pain in one's environment may be more associated with brain regions that are associated with affective and motor-motivational processes (de Waal and Preston, 2017). Furthermore, other factors such as the bodily location of pain, (e.g. face vs. limb) or laterality were not considered, but their importance has been demonstrated in previous metaanalyses (Jauniaux et al., 2019). On the other hand, the decision to include studies which depicted facial expressions of persons experiencing pain, rather than solely observation of pictures or video clips of directly delivered pain stimuli, could add an element of cognitive interpretation (Jauniaux et al., 2019;Xiong et al., 2019) which could also introduce some variance into the metaanalysis and should be considered as a limitation. Therefore, the present analysis should primarily be interpreted in relation to empathy for observation of pain through visual modality, and as a novel comparison with direct pain processing.
To surmise the impact of the present study, elucidation of a rich functional brain network of empathy for pain, extending beyond AI and aMCC, is important for theoretical understanding of the phenomenon. The conjunction and contrast analyses reveal, for the first time, shared and distinct networks for observed and direct pain which supports the concept of tiered levels of processing of empathy as were previously theorized (de Waal, 2008;Bernhardt and Singer, 2012;Lamm et al., 2016;de Waal and Preston, 2017). The findings indicate that an extended network of brain regions warrant greater focus and consideration for their role in empathic processing, particularly regions neighbouring TPJ in inferior parietal and occipitotemporal cortices. Such a broader approach would accord with recent developments and understanding from social neurosciences (Carter and Huettel, 2013;Igelström and Graziano, 2017). Finally, greater understanding of neurobiological underpinnings of empathy has practical implications for diagnosis of clinical disorders as well as therapeutic or pharmacological interventions (Decety et al., 2016;Christov-Moore and Iacoboni, 2019), or development of strategies to promote prosocial behaviours or establish moral, fair and practical societal structures (Decety and Cowell, 2015).
To conclude, the findings reveal concordance in an extensive bilateral network of brain regions for empathy for observed pain. This encompassed bilateral AI, supramarginal gyri, lateral occipitotemporal cortices IFG and aMCC, regions with functional relevance for interoception, pain processing, social cognition and self-other distinction. Utilizing novel contrast analyses for empathy for pain and direct pain experience, we demonstrated a broad network of shared brain representations which align to automaticity of response or emotional contagion, and empathy-specific activation patterns with relevance for higher order responses such as self-other distinction. Knowledge of these shared and distinct brain networks offers a novel insight into the neurobiological underpinnings of our subjective experience of empathy with relevance for theoretical, clinical and social applications.

Supplementary data
Supplementary data is available at SCAN online.

Conf lict of interest
None declared.