Introduction

Moral behavior is a building block of human societies and has classically been thought to be based on rational consideration. Aristotle (fourth century BC/1985), for example, argued that being a “good person” requires reasoning about virtues. Kant’s (1785/1993) famous categorical imperative was similarly rational, demanding that one should act according to principles that could also become a general law. More recently, Kohlberg et al. (1983) and Kohlberg (1969) advanced a six-stage developmental model acknowledging children’s increasing ability for abstraction and role-taking capacities in moral decisions. In contrast to rational models, the role of emotion in facilitating moral behavior has been less often emphasized (Haidt 2001). Hume (1777/2006) provides an early notable exception, as he believed in a key role of intuition for recognizing morally good and bad decisions, not requiring willful abstract reasoning. Charles Darwin (1874/1997) further argued that moral decisions are mainly influenced by emotional drives, which are rooted in socio-emotional instincts already present in non-human primates.

Notions of rationality and emotionality also serve as explanations in the contemporary neuroscientific literature on the psychological processes underlying moral decisions (henceforth: moral cognition). Results from functional magnetic resonance imaging (fMRI) studies by Greene et al. (2004) and Greene et al. (2001) were interpreted as revealing a dissociation between fast emotional responses and subsequent cognitive modulations in moral cognition. FMRI findings by Moll and Schulkin (2009), Moll et al. (2005a) and Moll et al. (2006), however, were interpreted as revealing a dissociation between group-oriented (i.e., pro-social) and self-oriented (i.e., egoistic) affective drives in moral cognition. Taken together, either abstract-inferential or rapid-emotional processing has been emphasized by most previous accounts explaining moral behavior. Rational explanations assumed that moral behavior arises from a conscious weighing of different rules, norms and situational factors, while emotional explanations emphasized the influence of uncontrolled emotional states rapidly evoked by a given situation (cf. Krebs 2008).

Previous evidence suggests that, the rational and emotional facets of moral cognition are likely related to other social skills: theory of mind (ToM) and empathy. ToM refers to the ability to contemplate other’s thoughts, desires, and behavioral dispositions by abstract inference (Premack and Woodruff 1978; Frith and Frith 2003). Indeed, accumulating evidence indicates that moral cognition is influenced by whether or not an agent’s action is perceived as intentional versus accidental (Knobe 2005; Cushman 2008; Killen et al. 2011; Moran et al. 2011). Empathy, on the other hand, refers to automatically adopting somebody’s emotional state while maintaining the self–other distinction (Singer and Lamm 2009; Decety and Jackson 2004). In moral decisions, experiencing empathy was shown to alleviate harmful actions towards others (Feshbach and Feshbach 1969; Zahn-Waxler et al. 1992; Eisenberger 2000). Conversely, the deficient empathy skills in psychopathic populations are believed to contribute to morally inappropriate behavior (Hare 2003; Blair 2007; Soderstrom 2003). Taken together, moral cognition is thought to involve the representation of intentions and outcomes as well as (vicariously experienced) emotional states (Decety et al. 2012; Leslie et al. 2006; Pizarro and Bloom 2003). This assumption is further supported by the observation that ToM and empathic skills precede mature moral reflection in primate evolution (Greene and Haidt 2002) and in ontogeny (Tomasello 2001; Kohlberg et al. 1983; Frith and Frith 2003; Piaget 1932) given that natural evolution tends to modify existing biological systems rather than create new ones from scratch (Jacob 1977; Krebs 2008).

Notably, theoretical accounts as well as empirical evidence suggest that ToM and empathy are partially overlapping psychological constructs. In particular, it has been proposed that embodied representations of affect, which should be relevant for empathic processing, may be further integrated into meta-representational or inferential processing (Keysers and Gazzola 2007; Spengler et al. 2009; De Lange et al. 2008; Mitchell 2005). That is, embodied representation and meta-representation might not constitute two mutually exclusive processes. In other words, more automatic, bottom-up driven mapping and awareness of others emotional states in the context of self–other distinction (i.e., empathy) might be modulated by more controlled, top-down processes involved in attributing mind states (i.e., ToM) (Leiberg and Anders 2006; Pizarro and Bloom 2003; Singer and Lamm 2009). Importantly, ToM and empathy differ in the representational content (mental states versus affect), yet both might be similarly brought about by interaction between top-down and bottom-up processes (Singer and Lamm 2009; Lamm et al. 2007; Spengler et al. 2009; Cheng et al. 2007). In line with the theoretical arguments for a partial overlap between ToM and empathy, ventromedial prefrontal cortex lesions associated with ToM impairments were empirically shown to debilitate elaborate forms of empathic processing, while lateral inferior frontal cortex lesions, which leave ToM skills intact, lead to an impairment of basic forms of empathic processing (Shamay-Tsoory et al. 2009). In sum, converging earlier evidence thus suggests that moral cognition is subserved by partially interrelated ToM and empathy processes.

From a neurobiological perspective, we therefore hypothesized that moral cognition might be subserved by brain areas also related to ToM and empathy, that is, brain networks associated with abstract-inferential and rapid-emotional processing, respectively. Furthermore, we sought to formally investigate to what extent the neurobiological correlates of ToM versus empathy overlap with the neural network associated with moral cognition. Moreover, we tested whether a common set of brain areas might be implicated in all three of these elaborate social-cognitive skills. We addressed these questions by means of a quantitative meta-analysis on peak coordinates reported in functional neuroimaging studies on moral cognition, ToM, and empathy.

Methods

Data used for the meta-analysis

We searched the PubMed database (http://www.pubmed.org) for fMRI and PET studies that investigated the neural correlates of moral cognition, ToM and empathy. Relevant papers were found by keyword search (search strings: “moral”, “morality”, “norm”, “transgression”, “violation”, “theory of mind”, “mentalizing”, “false belief”, “perspective taking”, “empathy”, “empathic”, “fMRI”, “PET”). Further studies were identified through review articles and reference tracing from the retrieved papers. Please note that in the context of ALE, the term “experiment” usually refers to any single (contrast) analysis on imaging data yielding localization information, while the term “study” usually refers to a scientific publication reporting one or more “experiments” (Laird et al. 2011; Eickhoff and Bzdok 2012). The inclusion criteria comprised full brain coverage as well as absence of pharmacological manipulations, brain lesions or mental/neurological disorders. Additionally, studies were only considered, if they reported results of whole-brain group analyses as coordinates corresponding to a standard reference space (Talairach/Tournoux, MNI). That is, experiments assessing neural effects in a priori defined regions of interest were excluded. We included all eligible neuroimaging studies published up to and including the year 2010. The exhaustive literature search yielded in the moral cognition category a total of 67 experiments reporting 507 activation foci; in the ToM category a total of 68 experiments reporting 724 activation foci, and in the empathy category a total of 112 experiments reporting 1,376 activation foci.

Methodologically equivalent to earlier ALE meta-analyses (Spreng et al. 2009; Lamm et al. 2011; Bzdok et al. 2011; Fusar-Poli et al. 2009; Mar 2011), study selection was grounded on the objective measure whether or not the authors claimed to have isolated brain activity that relates to moral cognition, ToM, or empathy. More specifically, we only included those neuroimaging studies into the “morality” category that required participants to make appropriateness judgments on actions of one individual towards others. In these studies, participants passively viewed or explicitly evaluated mainly textual, sometimes pictorial social scenarios with moral violations/dilemmas. The target conditions were frequently contrasted with neutral or unpleasant social scenarios (see Supplementary Table 1 for detailed study descriptions). Furthermore, we only included those neuroimaging studies into the “ToM” category that required participants to adopt an intentional stance towards others, that is, predict their thoughts, intentions, and future actions. These studies mostly presented cartoons and short narratives that necessitated understanding the beliefs of the acting characters. The target conditions were usually contrasted with non-social physical stories, which did not necessitate social perspective-taking. Finally, we only included those neuroimaging studies into the “empathy” category that aimed at eliciting the conscious and isomorphic experience of somebody else’s affective state. Put differently, in these studies participants were supposed to know and “feel into” what another person was feeling. These studies employed mostly visual, sometimes textual or auditory stimuli that conveyed affect-laden social situations which participants watched passively or evaluated on various dimensions.

Please note that we disregarded studies on empathy for pain because pain, although possessing an affective dimension, is not considered a classic emotion (Izard 1971; Ekman 1982). Rather, it is a bodily sensation mediated by distinct sensory receptors (Craig 2002; Saper 2000), and watching painful scenes does not induce isomorphic vicarious experiences (cf. Singer et al. 2004; Danziger et al. 2009). In particular, looking at a happy person usually elicits a sensation of happiness in the observer, yet watching a person in pain usually does not likewise evoke the physical experience of pain. Furthermore, we disregarded studies in which participants were presented with emotion recognition tasks using static pictures of emotional facial expressions. That is because such tasks are probably too simple to reliably entail sharing others’ emotions and maintaining a self–other distinction, both widely regarded as hallmarks of empathy (Singer and Lamm 2009).

Methodological foundation of activation likelihood estimation

The reported coordinates were analyzed for topographic convergence using the revised ALE algorithm for coordinate-based meta-analysis of neuroimaging results (Eickhoff et al. 2009; Turkeltaub et al. 2002; Laird et al. 2009a). The goal of coordinate-based meta-analysis of neuroimaging data is to identify brain areas where the reported foci of activation converge across published experiments. To this end, the meta-analysis determines if the clustering is significantly higher than expected under the null distribution of a random spatial association of results from the considered experiments while acknowledging the spatial uncertainty associated with neuroimaging foci.

As the first step, reported foci were interpreted as centers for 3D Gaussian probability distributions that capture the spatial uncertainty associated with each focus. This uncertainty is mostly a function of between-template (attributable to different normalization strategies and templates across laboratories) and between-subject (due to small sample sizes) variance. In fact, the between-template and between-subject variability are acknowledged based on empirical estimates, the latter being additionally gauged by individual sample size (Eickhoff et al. 2009).

In a second step, the probabilities of all activation foci in a certain experiment were combined for each voxel, yielding a modeled activation (MA) map (Turkeltaub et al. 2011). Voxel-wise ALE scores resulted from the union across these MA maps and quantified the convergence across experiments at each particular location in the brain.

The third and last step distinguished between random and ‘true’ convergence by comparing the ensuing ALE scores against an empirical null distribution reflecting a random spatial association between the experiments’ MA maps (Eickhoff et al. 2012). The within-experiment distribution of foci, however, was regarded to be fixed (Eickhoff et al. 2009). Thus, a random-effects inference was invoked, focusing on the above-chance convergence across different experiments (Eickhoff et al. 2009; Caspers et al. 2010; Kurth et al. 2010). The resulting ALE scores were tested against the earlier calculated ‘true’ ALE scores and cut off at a cluster-level-corrected threshold of p < 0.05. For cluster-level correction, the statistical image of uncorrected voxel-wise p values was first cut off by the cluster-forming threshold. Then, the size of the supra-threshold clusters was compared against a null distribution of cluster sizes derived from simulating 1,000 datasets with the same properties (number of foci, uncertainty, etc.) as the original experiments but random location of foci. The p value associated with each cluster was then given by the chance of observing a cluster of the given size in any particular simulation.

Additional conjunction and difference analyses were conducted to explore how different meta-analyses relate to each other. Conjunction-analyses testing for convergence between different meta-analyses employed inference by the minimum statistic, i.e., computing the intersection of the thresholded Z-maps (Caspers et al. 2010). That is, any voxel determined to be significant by the conjunction analysis constitutes a location in the brain which survived inference corrected for multiple comparisons in each of the individual meta-analyses. Difference analyses calculated the difference between corresponding voxels’ ALE scores for two sets of experiments. Then, the experiments contributing to either analysis were pooled and randomly divided into two analogous sets of experiments (Eickhoff et al. 2011). Voxel-wise ALE scores for these two sets were calculated and subtracted from each other. Repeating this process 10,000 times yielded a null distribution of recorded differences in ALE scores between two sets of experiments. The ‘true’ difference in ALE scores was then tested against these differences obtained under the null distribution yielding voxel-wise p values for the difference. These resulting non-parametric p values were thresholded at p < 0.001. Unfortunately, a statistical method to correct for multiple comparisons when assessing the differences between ALE maps has not yet been established. It should be mentioned, however, that the randomization procedure employed to compute the contrast between ALE-analyses is in itself highly conservative as it estimates the probability for a true difference between the two datasets.

Functional characterization

The converging activation patterns of tasks requiring moral cognition, theory of mind, or empathy were first determined by ALE meta-analysis. The conjunction across all three individual ALE analyses then yielded a computationally derived seed region for functional characterisation through quantitative correspondence with cognitive and experimental descriptions of the BrainMap taxonomy. In fact, BrainMap metadata describe the category of mental processes isolated by the statistical contrast of each experiment stored in the database (http://www.brainmap.org; Laird et al. 2009a). More specifically, behavioral domains (BD) include the main categories cognition, action, perception, emotion, interoception, as well as their subcategories. The respective paradigm classes (PC) categorize the specific task employed (a complete list of BDs and PCs can be found at http://www.brainmap.org/scribe/). We analyzed the behavioral domain and paradigm class metadata of BrainMap experiments associated with seed voxels to determine the frequency of domain ‘hits’ relative to its likelihood across the entire database. Using a binomial test, Bonferroni-corrected for multiple comparisons, the functional roles of the convergent network as a whole were identified by determining those BDs and PCs that showed a significant over-representation in experiments activating within the seed regions, relative to the entire BrainMap database (Laird et al. 2009b; Eickhoff et al. 2011).

Results

Individual meta-analyses of moral cognition, theory of mind, and empathy

All areas resulting from the ALE meta-analyses were anatomically labeled by reference to probabilistic cytoarchitectonic maps of the human brain using the SPM Anatomy Toolbox (see Tables 1, 2; Eickhoff et al. 2007; Eickhoff et al. 2005). Meta-analysis of neuroimaging studies related to moral cognition yielded convergent activation in the bilateral ventromedial/frontopolar/dorsomedial prefrontal cortex (vmPFC/FP/dmPFC), precuneus (Prec), temporo-parietal junction (TPJ), posterior cingulate cortex (PCC), as well as the right temporal pole (TP), right middle temporal gyrus (MTG), and left amygdala (AM) (Fig. 1; Table 1). Meta-analysis of studies related to ToM revealed convergence in the bilateral vmPFC/FP/dmPFC, Prec, TPJ, TP, MTG, posterior superior temporal sulcus (pSTS), and inferior frontal gyrus (IFG), as well as the right MT/V5. Finally, meta-analysis of studies related to empathy yielded convergence in the bilateral dmPFC, supplementary motor area (SMA), rostral anterior cingulate cortex (rACC), anterior mid-cingulate cortex (aMCC; cf. Vogt 2005), PCC, anterior insula (AI), IFG, midbrain, and TPJ, as well as the anterior thalamus on the left, further, the AM, MTG, pSTS, posterior thalamus, hippocampus, and pallidum on the right.

Table 1 Peaks of activations for the brain areas consistently engaged in fMRI studies on moral cognition, theory of mind, and empathy as revealed by ALE meta-analysis
Full size table
Table 2 Conjunction analyses that test for topographical convergence between the individual ALE meta-analyses of moral cognition, theory of mind, and empathy
Full size table
Fig. 1
figure 1

ALE meta-analysis of neuroimaging studies on moral cognition, theory of mind, and empathy. Significant meta-analysis results displayed on frontal, right, and left surface view as well as sagittal, coronal, and axial sections of the MNI single-subject template. Coordinates in MNI space. All results were significant at a cluster-forming threshold of p < 0.05 and an extent threshold of k = 10 voxels (to exclude presumably incidental results)

Full size image

Conjunction analyses

We conducted three conjunction analyses to examine where regions consistently involved in moral cognition converged with regions consistently involved in ToM, empathy, or both (Fig. 2; Table 2). In particular, we computed the conjunction across the individual meta-analysis of moral cognition and the difference between the analyses of ToM and empathy. Employing the difference excluded brain activity shared by both ToM and empathy. The conjunction across the neural network linked to moral cognition and the activation more robustly linked to ToM versus empathy revealed an overlap in the bilateral vmPFC/FP/dmPFC and TPJ, as well as the right MTG and TP. Conversely, the conjunction across the neural network linked to moral cognition and the activation more robustly linked to empathy versus ToM revealed an overlap in the dmPFC. Besides results from this conjunction analysis, it is noteworthy that the left AM was found in the analysis of moral cognition-related brain activity, while the right AM was found in the analysis of empathy-related brain activity. Moreover, the PCC showed significant convergence in both these individual meta-analyses at adjacent, yet non-overlapping locations. Specifically, convergence in the PCC was located slightly more rostro-dorsally in the ALE on moral cognition compared to the ALE on empathy (Fig. 1; Table 1). Finally, the conjunction analysis across the results of all three individual meta-analyses on moral cognition, ToM, and empathy yielded convergence in the dmPFC, right MTG, and bilateral TPJ. Additionally, we provide a list of those papers included in the present meta-analysis, which gave rise to the four converging clusters (Supplementary Table 2).

Fig. 2
figure 2

Conjunction analyses for topographical convergence across brain activity related to moral cognition and theory of mind (ToM) or empathy. Left panel overlapping activation patterns between the meta-analysis on moral cognition and the difference analysis between ToM and empathy (cluster-forming threshold: p < 0.05). Right bottom panel overlapping activation patterns between the meta-analysis on moral cognition and the difference analysis between empathy and ToM (cluster-forming threshold: p < 0.05). Right top panel sagittal and coronal slices of juxtaposed results from the meta-analyses on moral cognition (green) and empathy (red) to highlight similar convergence in the posterior cingulate cortex and amygdala (extent threshold: k = 10 voxels to exclude presumably incidental results). Coordinates in MNI space

Full size image

Functional characterization

Using the metadata of the BrainMap database for functional characterization as described above, we found that the common network observed across all three task domains was selectively associated with tasks tapping ToM, semantic processing, imagination, and social cognition, as well as episodic and explicit memory (Fig. 3). The network thus appears to be exclusively involved in higher-level cognitive processing.

Fig. 3
figure 3

Functional characterization of the core-network implicated in moral cognition, theory of mind (ToM), and empathy. Left neural network consistently activated across individual meta-analyses on moral cognition, ToM, and empathy (extent threshold: k = 10 voxels to exclude presumably incidental results). Images were rendered using Mango (multi-image analysis GUI, Research Imaging Institute, San Antonio, Texas, USA; http://ric.uthscsa.edu/mango/). Right functional characterization of the convergent network across all three tasks by BrainMap metadata. The purple bars denote the number of foci for that particular metadata class within the seed network. The grey bars represent the number of foci that would be expected to hit the particular seed network if all foci with the respective class were randomly distributed throughout the cerebral cortex. That is, the grey bars denote the by-chance frequency of that particular label given the size of the cluster. All shown taxonomic classes reached significance according to a binomial test (p < 0.05). Asterisks denote classes that survived the Bonferroni correction for multiple comparisons

Full size image

Discussion

Classic accounts of the psychology of moral cognition can be broadly divided into views that emphasize the involvement of either ‘rational’ or ‘emotional’ processes. In this study, we revisited this distinction by determining the consistent overlap between brain activation patterns reported in the neuroscientific literature on moral cognition with those of ToM (assumed to be more rational) and empathy (assumed to be more emotional) using coordinate-based ALE meta-analysis. By doing so, we demonstrated that moral cognition, indeed, recruits brain areas that are also involved in abstract-inferential (ToM) and rapid-emotional (empathy) social cognition. Furthermore, TPJ, mPFC, and MTG emerged as potential nodes of a network common to moral cognition, ToM, and empathy.

Individual analyses on the neural correlates of moral cognition, theory of mind, and empathy

To our knowledge, we here conducted the first quantitative meta-analytic assessment of the neural network engaged in moral cognition, that is, reflection of the social appropriateness of people’s actions. The obtained pattern of converged brain activation is in very good agreement with qualitative reviews of fMRI studies on moral cognition (Moll et al. 2005b; Greene and Haidt 2002). Concurrently, dysfunction in prefrontal, temporal and amygdalar regions is discussed as linked to psychopathy and anti-social behavior based on neuropsychological, lesion, and fMRI studies (Blair 2007; Anderson et al. 1999; Mendez et al. 2005).

The results of the meta-analysis of ToM were consistent with earlier quantitative analyses (Spreng et al. 2009; Mar 2011) of neuroimaging studies, in which participants attributed mental states to others to predict or explain their behavior. Put differently, the obtained network is likely to be implicated in the recognition and processing of others’ mental states (Amodio and Frith 2006; Gallagher and Frith 2003; Schilbach et al. 2010).

The meta-analysis on empathy across various affective modalities excluding pain was based on neuroimaging studies in which participants understood and vicariously shared the emotional experience of others. This analysis revealed the aMCC extending into the SMA and the AI extending into the IFG as the most prominent points of convergence. This is in line with a recent image-based meta-analysis on empathy for pain (Lamm et al. 2011). The present analysis on non-pain empathy, however, additionally revealed activation in the AM, rACC, and PCC. Consequently, empathy for pain- and non-pain-related affect appears to be implemented by an overlapping network that might recruit supplementary areas/networks depending on the specific affective modality with which participants are to empathize. In other words, the observed network is likely to be involved in vicariously experiencing others’ affective states (Singer et al. 2004; Wicker et al. 2003).

It is noteworthy that all brain areas revealed by the meta-analysis on moral cognition also converged significantly in the analyses on either ToM or empathy. It is thus tempting to speculate that moral cognition might rely on remodeling mental states and processing affective states of other people. Conversely, some of the brain areas, which show significant convergence in the ALE on ToM and empathy, were not significant in the ALE on moral cognition. This suggests that moral cognition might be realized by specific subsets, rather than the entirety of the neural correlates of ToM and empathy.

Converging neural correlates across moral cognition and theory of mind

Brain activity during moral cognition and ToM overlapped in the bilateral vmPFC/FP/dmPFC and TPJ, as well as the right TP and MTG. This extensive convergence indicates that moral cognition and ToM engage a highly similar neural network, which, in turn, entices to speculate about a close relationship between these two psychological processes. More specifically, increased activity along the dorso-ventral axis of the medial prefrontal cortex is heterogeneously discussed to reflect cognitive versus affective processes (Shamay-Tsoory et al. 2006), controlled/explicit versus automatic/implicit social cognition (Lieberman 2007; Forbes and Grafman 2010), goal versus outcome pathways (Krueger et al. 2009), dissimilarity versus similarity to self (Mitchell et al. 2006) as well as other-focus versus self-focus (Van Overwalle 2009). Apart from that, the spatially largest convergence across moral cognition- and ToM-related brain activity in the bilateral posterior temporal lobe/angular gyrus might be surprising given the divergent interpretation in the literature. That is, activation in this cortical region is conventionally interpreted as “posterior superior temporal sulcus” in the morality literature (Moll et al. 2005b; Greene and Haidt 2002) and as “temporo-parietal junction” in the ToM literature (Decety and Lamm 2007; Van Overwalle 2009; Saxe and Kanwisher 2003). This convergent activation in the ALE on moral cognition is, however, more accurately located to the TPJ, rather than to the pSTS (cf. Raine and Yang 2006; Binder et al. 2009). The potentially inconsistent neuroanatomical labeling might have disadvantageously affected discussion of this brain area in previous neuroimaging studies on moral cognition.

The engagement of ToM-associated areas during moral cognition has also been addressed in several recent fMRI studies by Young, Saxe, and colleagues. These authors proposed that during moral cognition, the dmPFC might process belief valence, while the TPJ and Prec might encode and integrate beliefs with other relevant features (Young and Saxe 2008). Especially, brain activity in the right TPJ was advocated to reflect belief processing during moral cognition. This argument was based on interaction effects with moral reasoning (Young et al. 2007), correlation with the participants’ self-reported tendency for acknowledging belief information (Young and Saxe 2009), and a significantly reduced impact of intentions after transient TPJ disruption (Young et al. 2010b). In line with our results, these fMRI studies suggest that moral cognition might involve reconstructing attributes and intentions that we apply to others and vice versa.

However, earlier neuroimaging evidence concerning the likely implication of the ToM network in moral cognition has two weak points. First, the interpretations of relevant fMRI studies were largely driven by reverse inference, i.e., identifying psychological processes from mere topography of brain activity (Poldrack 2006). That is, rather than investigating moral cognition tasks alone, the neural networks of moral cognition and ToM should have been compared directly by independent conditions involving either task. Second, the relevant fMRI studies crucially hinge on the repeatedly criticized functional localizer technique, which rests on strong a priori hypotheses (Friston et al. 2006; Mitchell 2008). The present ALE meta-analysis overcomes these two limitations by a largely hypothesis-free assessment of two independent pools of numerous whole-brain neuroimaging studies, providing strong evidence for the high convergence across the neural networks associated with moral cognition and ToM.

Converging neural correlates across moral cognition and empathy

Brain activity related to both moral cognition and non-pain empathy converged significantly in an area of the dmPFC, which was not revealed by the ALE on ToM. Converging activation of the dmPFC may suggest an implication of this highly associative cortical area in more complex social-emotional processing. In line with this interpretation, a recent fMRI study identified a similar brain location as highly selective for processing guilt (Wagner et al. 2012), an emotion closely related to moral and social transgression (Tangney et al. 2007). Moreover, the dmPFC has consistently been related to the (possibly interwoven) reflection of own and simulation of others’ mind states (Lamm et al. 2007; Jenkins and Mitchell 2010; Mitchell et al. 2006; D’Argembeau et al. 2007). Nevertheless, we feel that it might be currently unwarranted to confer precise functions to circumscribed parts of the dmPFC (cf. above), given the danger of reverse inference when deducing mental functions or states from regional activation patterns (Poldrack 2006). We therefore cautiously conclude that the observed convergence in the dmPFC probably reflects an unidentified, yet to be characterized, higher-level neural process that is related to affective and social processing.

It is interesting to note that the AM was significantly involved in the individual ALEs on morality and non-pain empathy, although in contralateral hemispheres. This brain region is thought to automatically extract biological significance from the environment (Ball et al. 2007; Sander et al. 2003; Bzdok et al. 2011; Müller et al. 2011). In particular, AM activity typically increases in the left hemisphere in elaborate social-cognitive processes and in the right hemisphere in automatic, basic emotional processes (Phelps et al. 2001; Bzdok et al. 2012; Markowitsch 1998; Gläscher and Adolphs 2003). This lateralization pattern thus seems well in line with the consistent engagement of the left AM in moral cognition and right AM in non-pain empathy. Moreover, activity in the PCC was found in adjacent, yet non-overlapping, locations during moral cognition and non-pain empathy. The PCC is thought important for the modality-independent retrieval of autobiographical memories and their integration with current emotional states (Fink et al. 1996; Maddock et al. 2001; Maddock 1999; Schilbach et al. 2008a). It might therefore be speculated that moral judgments and empathic processing could both rely on the integration on past experiences. Given that both PCC clusters do not overlap, however, it remains to be investigated whether (a) moral cognition and non-pain empathy engage distinct regions in the PCC, (b) the observed topographic pattern is purely incidental given the limited spatial resolution of meta-analyses, or (c) the differences in PPC activation reflect differences in stimulus-material and hence autobiographic associations.

Furthermore, it is important to note that the AI/IFG and aMCC/SMA were revealed as the most significant points of convergence in the ALE on non-pain empathy but did not show any overlap with the ALE on morality. In particular, the AI/IFG and aMCC/SMA form a network that is widely believed to represent one’s own and others’ emotional states regardless of the actual affective or sensory modality (Lamm et al. 2011; Wicker et al. 2003; Singer et al. 2004; Lamm and Singer 2010; Fan et al. 2011). Moreover, this network, especially the anterior insula, is not only implicated in meta-representation of emotional states but also in interoceptive awareness (Craig 2002, 2009; Kurth et al. 2010). Interoception-related meta-representation of emotion has thus been suggested to underlie the concomitant involvement of the aMCC/SMA and AI/IFG in neuroimaging studies on empathy as reflecting affective sharing (Singer and Lamm 2009; Fan et al. 2011).

In summary, we draw three conclusions from these observations. First, affective sharing, one core aspect of empathy, is unlikely to be involved in moral judgments, given the lack of consistent involvement of the AI/IFG or aMCC/SMA in paradigms probing the latter. Second, general affective processes might play a role in both moral judgments and empathy, given that the respective meta-analyses individually revealed the dmPFC (direct overlap) and the amygdala (oppositely lateralized). Third, the neural correlates of moral judgments are much closer related to the neural correlates of ToM than to those of empathy, as evidenced by the quantity of overlap in the conjunction analyses.

Converging neural correlates across all individual analyses

Moral cognition-, ToM-, and empathy-related brain activity converged in the bilateral TPJ, dmPFC, and right MTG, which therefore form a common network potentially involved in social-cognitive processes. In line with this, metadata profiling demonstrates solid associations of this network with neuroimaging studies related to ToM, explicit memory retrieval, language, and imagination of objects/scenes. Intriguingly, these four seemingly disparate psychological categories summarize what sets humans probably apart from non-human primates (Tomasello 2001; Frith and Frith 2010). They might functionally converge in the reciprocal relationship between the allocentric and egocentric perspective, instructed by self-reflection, social knowledge, and memories of past experiences (of social interactions). In particular, autobiographical memory supplies numerous building blocks of social semantic knowledge (Bar 2007; Binder et al. 2009). These isolated conceptual scripts may be reassembled to enable the forecasting of future events (Tulving 1983, 1985; Schacter et al. 2007). Similar brain mechanisms in remembering past episodes and envisioning future circumstances is supported by their engagement of identical brain areas, as evidenced by a quantitative meta-analysis (Spreng et al. 2009). Moreover, retrograde amnesic patients were reported to be impaired not only in prospection but also in imagining novel experience (Hassabis et al. 2007). These findings suggest a single neural network for mentally constructing plausible semantic scenarios of detached situations regardless of temporal orientation (Buckner and Carroll 2007; Hassabis and Maguire 2007). Indeed, construction of detached probabilistic scenes has been argued to influence ongoing decision making by estimating outcomes of behavioral choices (Boyer 2008; Suddendorf and Corballis 2007; Schilbach et al. 2008b). Taken together, moral cognition, ToM, and empathy jointly engage a network that might be involved in the automated prediction of social events that modulate behavior.

Relation to clinical research

Consistent with the demonstrated functional dissociation between cognitive and affective subsystems of the neural network related to moral cognition observed in our study, frontotemporal dementia has been reported to impair personal but not impersonal moral reasoning (Mendez et al. 2005). A cognitive-affective dissociation of moral cognition is also supported clinically by the psychopathic population’s immoral behavior in everyday life despite excellent moral reasoning skills (Cleckley 1941; Hare 1993). Apart from that, neither Greene’s nor Moll’s concept can exhaustively explain why vmPFC patients demonstrated a rationally biased approach to solving moral dilemmas (Koenigs et al. 2007; Moretto et al. 2010), yet, an emotionally biased approach to moral cognition in an economic game (Koenigs and Tranel 2007). Given the considerable neural commonalities of moral cognition and ToM tasks as revealed by the present analysis, contradictory findings of vmPFC patients dealing with moral dilemmas and economic games (Greene 2007; Koenigs and Tranel 2007; Koenigs et al. 2007) might resolve when meticulously probing ToM capabilities in future lesion studies.

Limitations

Several limitations of our study should be addressed. Meta-analyses are necessarily based on the available literature and hence may be affected by the potential publication-bias disfavoring null results (Rosenthal 1979). This caveat especially applies to the functional characterization of converging activation foci using the BrainMap database, as this database only contains about 21% of the published neuroimaging studies and its content therefore does not constitute a strictly representative sample. Furthermore, a part of the included neuroimaging studies on moral cognition might suffer from limited ecological validity. That is, the experimental tasks used might only partially involve the mental processes guiding real-life moral behavior (cf. Cima et al. 2010; Knutson et al. 2010; Young et al. 2010a). In particular, employing overly artificial moral scenarios (e.g., trolley dilemma), on top of the inherent limitations of neuroimaging paradigms, could have systematically overestimated cognitive versus emotional processes. This caveat might have contributed to the similar results in the meta-analyses on moral cognition and on ToM (rather than on empathy), given that emotion processing is thought to play a paramount role in real-world moral cognition (Krebs 2008; Tangney et al. 2007; Haidt 2001). Future neuroimaging studies should therefore strive for using more realistic moral scenarios to minimize the risk of investigating the neural correlates of “in vitro moral cognition” (cf. Schilbach et al. 2012; Schilbach 2010).

Conclusions

It is a topic of intense debate whether social cognition is subserved by a unitary specialized module or by a set of general-purpose mental operations (Mitchell 2006; Spreng et al. 2009; Bzdok et al. 2012; Van Overwalle 2011). The present large-scale meta-analysis provides evidence for a domain-global view of moral cognition, rather than for a distinct moral module (Hauser 2006), by showing its functional integration of distributed brain networks. More specifically, we parsed the neural correlates of moral cognition by reference to a socio-cognitive framework, exemplified by ToM cognition, and a socio-affective framework, exemplified by empathy. Ultimately, our results support the notion that moral reasoning is related to both seeing things from other persons’ points of view and to grasping others’ feelings (Piaget 1932; Tomasello 2001; Decety et al. 2012).