Abstract

A crucial feature of socially adaptive behavior is the ability to recognize when our actions harm other individuals. Previous research demonstrates that dorsal mediofrontal cortex (dMFC) and anterior insula (AI) are involved in both action monitoring and empathy for pain. Here, we tested whether these regions could integrate monitoring of error agency with the representation of others' pain. While undergoing event-related fMRI, participants played a visual task in turns with a friend placed outside the scanner, who would receive painful stimulation in half of the error trials. Brain activity was enhanced in dMFC and AI for painful compared with nonpainful errors. Left AI and dorsolateral pFC also exhibited a significant interaction with agency and increased responses when painful errors were caused by oneself. We conclude that AI is crucial for integrating inferences about others' feeling states with information about action agency and outcome, thus generating an affective signal that may guide subsequent adjustment.

INTRODUCTION

Perceiving oneself as the agent of actions and evaluating their consequences for other individuals are key elements for the control of social behavior. Understanding the affective states of others and avoiding harm to them lies at the root of moral emotions such as guilt and shame, which guide interactions between individuals and social groups. In cognitive neuroscience, action monitoring has been mainly studied in single-subject laboratory settings, typically focusing on errors in abstract cognitive tasks. Nothing is known about how the brain integrates the monitoring of one's own ongoing behavior with the evaluation of its social consequences, for example, when our errors cause loss or damage (e.g., pain) to another individual. Here we test whether the anterior insula (AI) and the dorsal mediofrontal cortex (dMFC)—two regions that have been implicated in both the representation of other persons' pain and error monitoring—could subserve this integrative function.

Abundant fMRI and electrophysiological research indicates that errors in cognitive tasks activate a specific network of regions in dMFC—ranging from dorsal anterior cingulate to SMA—as well as the dorsolateral pFC (dlPFC), inferior frontal gyrus (IFG), and AI (Klein et al., 2007; Taylor, Stern, & Gehring, 2007). It has been proposed that errors and task-related conflicts are initially detected by dMFC, with subsequent corrections in cognitive and motor control being implemented by dlPFC (Shackman et al., 2011; Ridderinkhof, Ullsperger, Crone, & Nieuwenhuis, 2004; Holroyd & Coles, 2002; Botvinick, Braver, Barch, Carter, & Cohen, 2001). In this framework, the role of the AI in error monitoring remains unclear. In line with studies highlighting its role in interoception and self-consciousness (Craig, 2009; Critchley, Wiens, Rotshtein, Ohman, & Dolan, 2004), it has been suggested that the AI might be critically involved in error awareness (Ullsperger, Harsay, Wessel, & Ridderinkhof, 2010; Klein et al., 2007). On the other hand, given its involvement in emotional states (Craig, 2009; Kober et al., 2008), the AI might be implicated in the affective component of error processing (Koban, Pourtois, Bediou, & Vuilleumier, 2012).

Indeed, in real life, errors may constitute emotionally salient and also socially significant events. However, only a few studies have examined the emotional and social components of action monitoring. For example, it has been found that observing another person committing an error may elicit partly similar brain responses to processing one's own action outcomes, as measured with EEG (Koban, Pourtois, Vocat, & Vuilleumier, 2010; Miltner, Brauer, Hecht, Trippe, & Coles, 2004; van Schie, Mars, Coles, & Bekkering, 2004) or fMRI (de Bruijn, de Lange, von Cramon, & Ullsperger, 2009; Shane, Stevens, Harenski, & Kiehl, 2008). Social factors can also modulate the monitoring of one's own, self-generated actions. Brain responses elicited by erroneous performance are amplified in cooperative settings where one's actions have direct effects on others (Koban et al., 2012; Radke, de Lange, Ullsperger, & de Bruijn, 2011), and this effect is stronger in participants with high trait empathy and perspective taking (Koban et al., 2012). Taken together, these findings suggest that social relationships may determine the salience of action outcome and their associated emotional value.

A crucial process to assess the impact of our actions on other people is the ability to represent the possible consequences for their mental and bodily states, for example, when our behavior may cause loss, injury, or pain to others. Several recent studies have revealed that seeing people in pain, or even just knowing that others receive pain, can activate brain regions responding to one's own pain, including AI and cingulate cortex, as well as periacqueductal gray, thalamus, and cerebellum (Corradi-Dell'Acqua, Hofstetter, & Vuilleumier, 2011; Lamm, Decety, & Singer, 2011; Ochsner et al., 2008; Jackson, Meltzoff, & Decety, 2005; Singer et al., 2004). Although these findings have been mainly discussed in terms of their relationship to neural representations shared between one's own and others' pain sensation, this network is also highly reminiscent of error monitoring and negative affect (Shackman et al., 2011; Ullsperger et al., 2010; Taylor et al., 2007).

Indeed, a quantitative meta-analysis based on the term “errors” and “empathy” in the NeuroSynth database (www.neurosynth.com; Yarkoni, Poldrack, Nichols, Van Essen, & Wager, 2011; see Methods) clearly points to common brain regions in a voxel-based way. As illustrated in Figure 2A, meta-analytic activations related to either error monitoring or empathy show highly significant conjunction effects in dMFC, but also in AI and IFG. This overlap between error monitoring and empathy (or more generally negative affect and nociception Shackman et al., 2011) points to the possibility of common neural pathways in dMFC and AI for these different processes. This convergence might underlie the aversive signals guarding against harmful actions toward others and the moral sense of action responsibility. Moreover, some authors proposed that the AI may directly contribute to the experience of agency for actions (Farrer & Frith, 2002), perhaps by integrating information about agency, intention, and valence of outcomes (Brass & Haggard, 2010). However, the integration of agency, pain, and action control has never been investigated. Furthermore, although both dMFC and AI show overlapping activations for pain and error monitoring, it remains unresolved whether these two areas subserve similar or distinct evaluative roles (Shackman et al., 2011).

Here, we designed a novel fMRI paradigm to investigate the monitoring of harmful errors, in which two participants were engaged in a cooperative game (Figure 1). One laid in the MRI scanner while his or her brain activity was recorded, whereas the other sat in front of a computer next to the scanner room. Both participants saw the same visual display and alternated between performing a difficult dot estimation task and observing the performance of the other participant in the same task. At the end of each trial, a feedback signaled to both participants whether the response was correct or erroneous. Critically, half of the error trials were associated with painful thermal stimulation to the person outside the scanner. This yielded, from the point of view of the subject inside the scanner, four different types of errors: when playing the game, his or her own error caused pain to the other (Playing-PainfulError), or caused no pain (Playing-NonpainfulError); whereas when observing, errors made by the other caused pain to him/herself (Observing-PainfulError) or caused no pain (Observing-NonpainfulError). Of crucial interest for our study is the interaction effect, [Playing-PainfulError > Playing-NonpainfulError] > [Observing-PainfulError > Observing-NonpainfulError], which allowed us to identify specific brain responses to those events where another person's pain was perceived as “my fault,” while controlling for effects associated with the representation of others' pain per se, and with error agency per se. In line with the meta-analysis findings (Figure 2A), we focused our hypotheses on AI and dMFC. If these regions hold separate neural processes responding to empathic pain and error agency independently from each other, we predict additive effects of these two factors. On the other hand, if empathic pain and error agency are integrated to signal the social and moral relevance of action outcomes, then these regions should show an interaction between these factors.

Figure 1. 

Experimental setup and design. (A) Two participants took turns either performing or observing the other's performance in a dot-counting task. Brain activity of one participant was recorded in a 3T-fMRI scanner, whereas the second participant was outside the scanning room and received either painful or nonpainful thermal stimulation after each trial, depending on the performance of the current player. A visual feedback about performance was always presented to both participants simultaneously. (B) The dot-counting task required the participants to indicate the side of the screen with the greater number of dots. (C) The full 2 × 3 factorial design resulting from the different feedback type in the two task conditions (playing or observing).

Figure 1. 

Experimental setup and design. (A) Two participants took turns either performing or observing the other's performance in a dot-counting task. Brain activity of one participant was recorded in a 3T-fMRI scanner, whereas the second participant was outside the scanning room and received either painful or nonpainful thermal stimulation after each trial, depending on the performance of the current player. A visual feedback about performance was always presented to both participants simultaneously. (B) The dot-counting task required the participants to indicate the side of the screen with the greater number of dots. (C) The full 2 × 3 factorial design resulting from the different feedback type in the two task conditions (playing or observing).

Figure 2. 

Brain activation to errors, empathy, and agency. (A) NeuroSynth meta-analytic forward inference maps for the terms errors (in red) and empathy (in green), thresholded at p < .05, FDR-corrected (Yarkoni, Poldrack, Nichols, et al., 2011). (B) Significant activations (p = .05 FEW-corrected at cluster level) are depicted on an inflated brain template for the contrasts of Error > Correct, Painful > Nonpainful Errors, and Playing > Observing.

Figure 2. 

Brain activation to errors, empathy, and agency. (A) NeuroSynth meta-analytic forward inference maps for the terms errors (in red) and empathy (in green), thresholded at p < .05, FDR-corrected (Yarkoni, Poldrack, Nichols, et al., 2011). (B) Significant activations (p = .05 FEW-corrected at cluster level) are depicted on an inflated brain template for the contrasts of Error > Correct, Painful > Nonpainful Errors, and Playing > Observing.

METHODS

NeuroSynth Meta-analysis

Meta-analyses compensate for some of the shortcomings inherent to functional imaging studies, such as small number of participants or inflated statistical thresholds because of multiple comparisons (Yarkoni, Poldrack, Van Essen, & Wager, 2011; Fox, Parsons, & Lancaster, 1998). NeuroSynth is an ambitious project aimed at automatic identification, extraction, and synthesis of human functional imaging results and corresponding metadata (Yarkoni, Poldrack, Nichols, et al., 2011). NeuroSynth uses text-mining techniques to detect frequently used terms (as proxies for concepts of interest) in the neuroimaging literature: Terms that occur at a high frequency in a given study are then associated with all activation coordinates in this publication, allowing for automated term-based meta-analysis. Compared with manual data entry, NeuroSynth has the advantage of automatic processing of a very large sample of studies (currently >4000 papers). Despite the automaticity and the potentially high noise resulting from the association between term frequency and coordinate tables, this approach has been shown to be quite robust and reliable (Yarkoni, Poldrack, Nichols, et al., 2011).

The NeuroSynth database was accessed on May 20, 2012, to characterize meta-analytic activation maps (forward inference) associated with the terms “empathy” (n = 39 studies) and “errors” (n = 211 studies). These maps were downloaded and are displayed in Figure 2A at a threshold of p < .05 (FDR-corrected; see www.neurosynth.com and Yarkoni, Poldrack, Nichols, et al., 2011, for more information).

Participants

Twenty-two volunteers (10 men, ranging from 18 to 33 years) took part in the study. The recruitment was organized so that participants were members of 11 pairs of the same sex and friends with one another. As two of the women recruited did not pass the MRI safety criteria, they did not undergo the scanning session but took part only in the physiological recording outside the scanner. Data from one female participant were excluded because of a history of affective disorders and psychotropic medication. The final fMRI sample thus consisted of 19 healthy adults (10 men and 9 women, mean age = 24.3 years). All participants provided informed consent and were paid for their participation. The study design was approved by the ethics committee of the University Hospital Geneva.

Stimuli and Task

Participants took turns performing and observing a dot-counting task, adapted from a previous study on social interactions (Vrticka, Andersson, Grandjean, Sander, & Vuilleumier, 2008). Both saw the exact same visual display. Each trial (Figure 1B) started with a central fixation cross and the name of the current player (presented for a randomized duration of 2–9 sec, with an exponentially decreasing distribution). Then two visual stimuli were presented for 500 msec: These were two randomly selected sets of 10–15 dots, shown on a gray background on the left and right side of the screen, separated by a central vertical white line. Participants had to indicate which side contained the largest amount of dots by pressing one of two keys with their right hand (index/middle finger). After a jittered interval of 1500–3500 msec, a visual feedback about performance was given (shown for 5000 msec). At the same time, thermal stimulation given to the participant outside the scanner was changed from baseline (34°C) to either warm but nonpainful temperature (40°C, raising slope 3°C/sec) or to hot and painful temperature (set by subject-specific tolerance threshold obtained in a prior session; average ∼ 44.7°C, slope 8°C/sec). The target temperature remained at the same level for 2 sec before returning to baseline (falling slope 6°C/sec). A simultaneous visual feedback was displayed and consisted of colored circles (equiluminant) containing simple geometrical shapes inside them (see Figure 1C). The colors could be either green or orange, respectively indicating that a correct response or an error was made by the player. The inside shapes could be a rectangle or a triangle, respectively indicating that either painless or painful thermal stimulation was given to the participant outside the scanner. Furthermore, each feedback display also contained (in the center of the colored circle) an indication of monetary gain/loss based on the trial outcome. These served as an additional motivation to maintain an optimal level of task performance and to ensure attention during both the playing and observing conditions. The amounts were +20 cents (correct response) and −10 cents (incorrect response), and participants were told that they would win or lose these monetary incentives together, as a team. Importantly, whereas the monetary outcome was only determined by the actual performance in the counting task (error or correct), the thermal stimulation was always painless for correct trials, but painful on 50% of the trials following an error (randomized and unpredictable order).

Trial difficulty was measured by the absolute difference between the numbers of dots in the two sides of the screen. Following our previous study (Vrticka et al., 2008), task difficulty (number of points) was adjusted on-line throughout the whole experiment according to the current participant's performance, with an upper limit of correct responses set to 56% of trials. This adaptive criterion was implemented to ensure a comparable amount of correct and incorrect trials for each participant. Participants were neither informed nor aware that the task difficulty was adapted over time (as confirmed by debriefing). The experiment was designed and controlled using E-Prime 2.0 (Psychology Software Tools, Inc.).

Procedure

Following informed consent and task instructions, the two participants of each pair were familiarized with the thermal stimulation device, and an individual pain tolerance threshold was determined for each person outside the magnet room, before the first functional imaging block. Following Corradi-Dell'Acqua et al. (2011), the subject-specific threshold was calculated by an ascending method of limits and corresponded to stimulation intensity sufficiently strong to be considered painful, but sufficiently weak to be felt without moving. The two participants then performed a short session (10 practice trials) of the dot-counting task while sitting next to each other, ensuring that both of them understood the task and the different types of outcomes. During this session, each in turn received the thermal stimulation during the three different feedback conditions (correct-nonpainful, error-nonpainful, error-painful); thus, both participants could directly experience the painful stimulation, enhancing the credibility of the experiment.

The whole experimental session was divided into four blocks, each consisting of 80 trials (approximately 16 min). In two blocks of four, the dot-counting performance of one member of the pair was recorded in the MRI scanner, whereas the second member performed the task on a PC station outside the scanner (Figure 1A). In the other two blocks, the participants switched roles. This ensured that half of participants underwent 32 min of behavioral testing first and then another 32 min of MRI testing, with the inverse order for the remaining half of participants. The only exceptions were those pairs in which one member did not undergo the scanning session because of MRI safety criteria (see above): in this case, two sessions were carried out, in which only the member complying with MRI safety criteria went inside the scanner.

The same experimental trials were presented to both participants (inside and outside MRI) at the same time. They both performed the dot-counting task in alternating turns. While one played, the second passively watched the trial, including the feedback information. Note that, irrespective of who played the task, the thermal stimulation associated with feedback was always administered to the person outside the scanner. Turns were organized as follows: 80 trials within one block were divided in 16 miniblocks of 5 trials each (starting order counterbalanced across sessions). Before each miniblock, a text string of 2.5-sec duration informed the participants on who had to play in the forthcoming trials.

Both participants watched the same visual display on two different but synchronized screens. Inside the scanner, this was achieved by a mirror mounted on the headcoil reflecting images sent by an LCD projector (CP-SX1350, Hitachi, Japan), with about 14.25° (vertical) × 19° of visual angle. Key presses were recorded with an MRI-compatible bimanual response button box (HH-2 × 4-C, Current Designs, Inc., USA).

Thermal Stimulation

We used MSA Thermotest equipment (SOMEDIC Sales AB, Sweden) to apply nonpainful and painful thermal stimulation. For experimental sessions, the thermode was placed on the left calf of the player in the scanner control room and fixed between the skin and trousers in such a way that it could be easily and quickly removed in case of unbearable discomfort (which however never happened during the study).

Behavioral and Personality Measures

After all task blocks were run, we asked participants to rate the intensity of empathic pain on a Likert scale (ranging from 1 to 10), as they experienced it overall on trials when they caused pain to their coplayers (caused painful errors) and when they observed painful feedback given to the other as a consequence of the coplayer own errors (observed painful errors). To investigate whether causing somebody else's pain was indeed associated with feelings of guilt and other negative emotions, we also asked participants to indicate the level of five negative emotions for each trial type (on a scale from 1 to 5): fear, shame, guilt, sadness, and anger.

Image Acquisition and Statistical Analyses

MRI images were acquired using a 3-T whole-body scanner (Trio TIM, Siemens, Germany) with a 12-channel head coil. The structural image of each participant was recorded with a T1-weighted MPRAGE sequence (repetition time = 1900 msec, inversion time = 900 msec, echo time = 2.27 msec, 1 × 1 × 1 mm voxel size). We applied a standard T2-weighted EPI sequence (2D-EP, repetition time = 2100 msec, echo time = 30 msec, flip angle = 80°, 3.2 × 3.2 × 3.2 mm voxel size) for acquisition of functional images of the whole brain (36 slices).

SPM8 software (Wellcome Department of Imaging Neuroscience, UCL, London, U.K.; www.fil.ion.ucl.ac.uk/spm) was used for image preprocessing and analyses. Preprocessing followed standard practice, including realignment and reslicing, coregistration of the functional images to the structural image, unified segmentation and normalization (resampling to 2 × 2 × 2 mm voxel size), and smoothing using an 8 × 8 × 8 mm Gaussian kernel. Eight regressors modeled the functional imaging data in a first-level analysis for each subject, including the onset of the dot stimuli for both the playing and observing conditions, plus the six different feedback conditions: Playing-Correct (PC), Observed-Correct (OC), Playing-NonpainfulError (PNP), Observed-NonpainfulError (ONP), Playing-PainfulError (PP), and Observed-PainfulError (OP). Time derivatives of these regressors were added to the model to correct for slice time differences as well as possible uncertainties in the timing of the psychological events of interest. Parametric modeling with a linear time modulation was included to account for habituation and order effects. Finally, six regressors of no interest were included to correct for movement artifacts. Data were high-pass filtered to reduce low-frequency noise (cutoff 128 sec). Estimation of parameters at each voxel used a restricted maximum likelihood with an autoregressive AR(1) model to account for temporal autocorrelation.

For the second-level group analysis, the six parameter estimates associated with each feedback condition were fed into a flexible factorial design with a within-subject factor of six levels using a random effects analysis in SPM8. In this group analysis, activations were considered significant if they survived a height threshold of t > 3.18 (corresponding to p < .001 uncorrected at the voxel level) and an extent threshold of k > 193 consecutive voxels, corresponding to p < .05 family-wise corrected at the cluster level for multiple comparisons across the whole brain.

RESULTS

Behavior

The distribution of different outcomes was comparable across the two agency conditions (playing vs. observing). Participants in the scanner responded correctly in 55.6% of the trials (SD = 5.6%), whereas painless errors were made on 21.6% (SD = 4.5%) of the trials and painful errors on 22.8% (SD = 6.1%). The coplayers outside the scanner performed highly similarly, yielding 56.0% (SD = 4.9%) of correct responses, 20.1% (SD = 5.6%) of painless errors, and 23.1% (SD = 4.6%) of painful errors.

After the experimental runs, an overall rating (Likert scale) was given by the participant in the scanner for the pain intensity thought to be experienced by the other person during painful errors, both when playing and when observing. These empathic ratings revealed slightly higher pain intensity when the player caused the pain of his partner (3.3 vs. 2.9, nonparametric sign test, Z = 2.0, p = .046).

Finally, subjective reports of emotions (fear, shame, guilt, anger, and sadness) experienced for different action outcomes while performing the task inside the scanner were also given by each participant after scanning (see Table 1). To control for nonspecific emotional effects because of self-generated errors, we calculated the difference for each emotion rating between the painful and nonpainful errors, in each agency condition (playing and observing). As expected, these data showed globally higher emotional responses to painful compared with nonpainful errors on all scales (Figure 3). Pairwise Bonferroni-corrected comparisons with sign tests revealed significant differences between the two agency conditions for the emotion of guilt (Z = 2.9, p = .003) only, although there was also an uncorrected effect for shame (p = .041). These results indicate that moral emotions were specifically induced when painful errors were “caused by me,” whereas other negative emotions were more generally elicited by error feedback and pain stimulation to the coplayer irrespective of agency. Taken together, these behavioral data show that monitoring the outcome of actions with painful consequences for another person triggers specific affective states in the agent and modulate his or her own empathic perception of the pain experienced by the other.

Table 1. 

Behavioral Results from the Emotion Ratings for the Different Trial Types (Means and SDs)


Playing [Mean (SD)]
Observing [Mean (SD)]
Corr
NP
P
Corr
NP
P
Fear 1.2 (0.9) 1.1 (0.3) 1.9 (1.2) 1.2 (0.9) 1.4 (1.1) 2.2 (1.4) 
Shame 1.0 (0.0) 2.2 (1.4) 3.2 (1.4) 1.0 (0.0) 1.1 (0.5) 1.2 (0.9) 
Guilt 1.0 (0.0) 2.5 (1.4) 4.1 (1.1) 1.0 (0.0) 1.0 (0.0) 1.3 (0.9) 
Sadness 1.0 (0.0) 1.9 (1.5) 2.2 (1.5) 1.0 (0.0) 1.3 (0.7) 1.9 (1.5) 
Anger 1.0 (0.0) 2.9 (1.4) 3.3 (1.8) 1.0 (0.0) 1.1 (0.3) 1.6 (1.0) 

Playing [Mean (SD)]
Observing [Mean (SD)]
Corr
NP
P
Corr
NP
P
Fear 1.2 (0.9) 1.1 (0.3) 1.9 (1.2) 1.2 (0.9) 1.4 (1.1) 2.2 (1.4) 
Shame 1.0 (0.0) 2.2 (1.4) 3.2 (1.4) 1.0 (0.0) 1.1 (0.5) 1.2 (0.9) 
Guilt 1.0 (0.0) 2.5 (1.4) 4.1 (1.1) 1.0 (0.0) 1.0 (0.0) 1.3 (0.9) 
Sadness 1.0 (0.0) 1.9 (1.5) 2.2 (1.5) 1.0 (0.0) 1.3 (0.7) 1.9 (1.5) 
Anger 1.0 (0.0) 2.9 (1.4) 3.3 (1.8) 1.0 (0.0) 1.1 (0.3) 1.6 (1.0) 

The five different emotions were rated on 1–5 Likert scales, after the main experimental blocks.

Figure 3. 

Emotion ratings for painful errors. The average intensity reported for different emotion categories is plotted as a function of feedback and agency, corrected for more general feedback effects by subtracting the ratings for nonpainful errors. Data are shown separately for playing and observing.

Figure 3. 

Emotion ratings for painful errors. The average intensity reported for different emotion categories is plotted as a function of feedback and agency, corrected for more general feedback effects by subtracting the ratings for nonpainful errors. Data are shown separately for playing and observing.

Functional Imaging

Main Effects

All regions surviving correction for multiple comparisons at the cluster level for the contrasts of main interest are displayed in Figure 2B. We first investigated the main effect of negative (error) versus correct performance feedback across agency and pain conditions (Table 2). Increased activation for errors compared with correct feedback was found in widespread regions across dMFC, including the anterior cingulate and medial superior frontal gyrus, in line with the meta-analytic maps (Figure 2A). Further activations were found bilaterally in AI, as well as in lateral prefrontal and temporo-occipital areas (Figure 2B, red clusters). Positive feedback, on the other hand, led to robust activation in the left and right ventral striatum (Table 2).

Table 2. 

Main Effect of Outcome (Error vs. Correct)

RegionsLeft/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Errors > Correct 
SMA, medial frontal gyrus, cingulate gyrus R/L 16 58 5251 7.62 
−4 24 54 6.99 
40 44 6.61 
IFG, frontal inferior operculum, insula, middle frontal gyrus 52 28 2776 7.54 
54 18 28 6.58 
52 26 18 6.11 
Middle occipital gyrus, fusiform gyrus 42 −70 −10 2230 6.85 
34 −84 12 6.58 
34 −64 −18 5.30 
Inferior temporal gyrus, middle occipital gyrus −48 −68 −6 1487 6.59 
−44 −82 −4 5.42 
−30 −92 16 5.13 
Insula, IFG, frontal inferior operculum −48 20 −8 2922 6.23 
−32 26 −6 6.19 
−40 22 5.54 
Thalamus, midbrain R/L −24 638 5.55 
−6 −24 −2 5.32 
12 −10 −6 3.99 
Cuneus, precuneus −10 −70 30 338 4.87 
−20 −64 26 4.12 
Middle temporal gyrus, supramarginal gyrus 50 −34 −2 636 4.79 
54 −44 28 4.62 
62 −38 4.52 
Posterior cingulate −8 −70 10 255 4.46 
 
Correct > Errors 
Caudate −12 −10 658 6.59 
−12 14 −4 4.97 
−18 24 14 4.94 
Caudate 10 −8 357 5.75 
16 22 3.93 
RegionsLeft/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Errors > Correct 
SMA, medial frontal gyrus, cingulate gyrus R/L 16 58 5251 7.62 
−4 24 54 6.99 
40 44 6.61 
IFG, frontal inferior operculum, insula, middle frontal gyrus 52 28 2776 7.54 
54 18 28 6.58 
52 26 18 6.11 
Middle occipital gyrus, fusiform gyrus 42 −70 −10 2230 6.85 
34 −84 12 6.58 
34 −64 −18 5.30 
Inferior temporal gyrus, middle occipital gyrus −48 −68 −6 1487 6.59 
−44 −82 −4 5.42 
−30 −92 16 5.13 
Insula, IFG, frontal inferior operculum −48 20 −8 2922 6.23 
−32 26 −6 6.19 
−40 22 5.54 
Thalamus, midbrain R/L −24 638 5.55 
−6 −24 −2 5.32 
12 −10 −6 3.99 
Cuneus, precuneus −10 −70 30 338 4.87 
−20 −64 26 4.12 
Middle temporal gyrus, supramarginal gyrus 50 −34 −2 636 4.79 
54 −44 28 4.62 
62 −38 4.52 
Posterior cingulate −8 −70 10 255 4.46 
 
Correct > Errors 
Caudate −12 −10 658 6.59 
−12 14 −4 4.97 
−18 24 14 4.94 
Caudate 10 −8 357 5.75 
16 22 3.93 

All activations are corrected for multiple comparisons across the whole brain, p < .05 FWE (cluster level).

The main effect of painful versus nonpainful errors (Table 3) also showed significant activations in dMFC, AI, and left IFG, in line with other studies on vicarious pain processing (Lamm et al., 2011; Figure 2A). Areas in AI and dMFC overlapped with the main effect of errors (see Figure 2B). Additional increases were observed in bilateral occipital areas, precuneus, and thalamus. No region was activated for nonpainful > painful errors at the same corrected threshold.

Table 3. 

Main Effect of Pain (Painful > Painless Error Feedback)

Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Middle occipital gyrus, inferior temporal gyrus 34 −84 10 2588 10.18 
46 −76 −6 6.87 
34 −64 −4 5.26 
Middle occipital gyrus, inferior occipital gyrus −36 −86 2198 8.37 
−44 −78 −6 7.93 
−28 −92 14 7.54 
Medial frontal gyrus, cingulate gyrus R/L −8 42 26 1345 6.14 
−8 48 20 5.08 
−4 16 32 4.69 
IFG, Insula −48 20 −4 752 5.18 
−30 24 −10 4.98 
−36 26 4.88 
Precuneus −14 −70 30 349 4.60 
Thalamus R/L −10 −4 423 4.21 
−6 −30 −2 4.06 
−24 4.03 
Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Middle occipital gyrus, inferior temporal gyrus 34 −84 10 2588 10.18 
46 −76 −6 6.87 
34 −64 −4 5.26 
Middle occipital gyrus, inferior occipital gyrus −36 −86 2198 8.37 
−44 −78 −6 7.93 
−28 −92 14 7.54 
Medial frontal gyrus, cingulate gyrus R/L −8 42 26 1345 6.14 
−8 48 20 5.08 
−4 16 32 4.69 
IFG, Insula −48 20 −4 752 5.18 
−30 24 −10 4.98 
−36 26 4.88 
Precuneus −14 −70 30 349 4.60 
Thalamus R/L −10 −4 423 4.21 
−6 −30 −2 4.06 
−24 4.03 

All activations are corrected for multiple comparisons across the whole brain, p < .05 FWE (cluster level).

We then assessed the main effect of agency, reflecting differences in the processing of feedback for one's own versus observed actions, across all three types of outcomes (correct, painless error, painful error). This comparison revealed significant activations in dMFC, bilateral dlPFC, and cerebellum (Figure 2B, blue clusters). These data suggest a general role of these regions in self-monitoring. The opposite contrast (observed > own feedback) yielded only a few clusters in caudate nucleus and precuneus.

Interaction Effects

Most critically, we next investigated how agency of the action influenced the processing of its outcome by looking at any interactions between Correctness (correct feedback vs. all errors) and Agency (playing vs. observing). An interaction testing for increased neural responses to one's own (but not other's) correct performance (i.e., in the contrast [Playing-Correct > Playing-Error] > [Observing-Correct > Observing-Error]) was selectively observed in the ventral striatum, plus the middle occipital gyrus, in both hemispheres (Table 4 and Figure 4A). As depicted in Figure 4A, this interaction in the striatum was driven by a markedly lower signal for errors (irrespective of painfulness) compared with correct trials when the participant was responsible for the outcome (playing condition), whereas no modulation was found in the observation condition.

Table 4. 

Interaction Effect Outcome × Agency

Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
[Playing-Correct > Playing-Error] > [Observed-Correct > Observed-Error] 
Ventral caudate, putamen 14 −6 638 7.29 
Ventral caudate, putamen −12 10 −6 254 5.96 
Middle occipital gyrus 24 −98 393 4.93 
20 −94 14 4.07 
Middle occipital gyrus −22 −94 417 4.48 
−12 −102 10 4.20 
−22 −88 10 3.61 
 
[Playing-Error > Playing-Correct] > [Observed-Error > Observed-Correct] 
Cingulate gyrus, medial frontal gyrus R/L −4 22 30 1651 6.46 
20 22 4.78 
−2 36 4.45 
SMA R/L −6 62 464 5.19 
10 16 64 4.24 
−2 24 56 3.96 
Rolandic operculum, right insula, postcentral gyrus 48 −8 16 438 4.48 
42 −16 20 4.38 
64 −6 34 3.98 
Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
[Playing-Correct > Playing-Error] > [Observed-Correct > Observed-Error] 
Ventral caudate, putamen 14 −6 638 7.29 
Ventral caudate, putamen −12 10 −6 254 5.96 
Middle occipital gyrus 24 −98 393 4.93 
20 −94 14 4.07 
Middle occipital gyrus −22 −94 417 4.48 
−12 −102 10 4.20 
−22 −88 10 3.61 
 
[Playing-Error > Playing-Correct] > [Observed-Error > Observed-Correct] 
Cingulate gyrus, medial frontal gyrus R/L −4 22 30 1651 6.46 
20 22 4.78 
−2 36 4.45 
SMA R/L −6 62 464 5.19 
10 16 64 4.24 
−2 24 56 3.96 
Rolandic operculum, right insula, postcentral gyrus 48 −8 16 438 4.48 
42 −16 20 4.38 
64 −6 34 3.98 

All activations are corrected for multiple comparisons across the whole brain, p < .05 FWE (cluster-level).

Figure 4. 

SPM results for interaction effects. (A) For the interaction of Correctness × Agency, selective effects were observed in the left and right ventral striatum. Parameter estimates of activity (beta values) averaged across voxels from these two clusters are depicted for each experimental condition. (B) Interaction effects of Error × Agency were observed in the dMFC and adjacent areas. Parameter estimates of activity (beta values) averaged across voxels from the dMFC cluster are depicted for each experimental condition. (C) For the interaction of Pain × Agency, effects were observed in the left AI and left dorsolateral prefrontal cortex. Parameter estimates of activity (beta values) averaged across voxels from these two clusters are depicted for each experimental condition. All activation maps are displayed at a threshold of p = .05, FEW-corrected at cluster level across the whole brain. Vertical bars indicate standard errors. Corr = correct trials; NP = nonpainful errors; P = painful errors.

Figure 4. 

SPM results for interaction effects. (A) For the interaction of Correctness × Agency, selective effects were observed in the left and right ventral striatum. Parameter estimates of activity (beta values) averaged across voxels from these two clusters are depicted for each experimental condition. (B) Interaction effects of Error × Agency were observed in the dMFC and adjacent areas. Parameter estimates of activity (beta values) averaged across voxels from the dMFC cluster are depicted for each experimental condition. (C) For the interaction of Pain × Agency, effects were observed in the left AI and left dorsolateral prefrontal cortex. Parameter estimates of activity (beta values) averaged across voxels from these two clusters are depicted for each experimental condition. All activation maps are displayed at a threshold of p = .05, FEW-corrected at cluster level across the whole brain. Vertical bars indicate standard errors. Corr = correct trials; NP = nonpainful errors; P = painful errors.

For the opposite interaction, highlighting greater responses to one's own errors relative to the other's (i.e., contrast [Playing-Error > Playing-Correct] > [Observing-Error > Observing-Correct], we found strong increases in the dMFC (Table 4 and Figure 4B), over and around the regions previously implicated in the main effect for errors and for vicarious pain (Figure 2). Figure 4B illustrates these dMFC effects in the different feedback conditions, showing not only that these regions exhibited a selective increase to errors as compared with correct feedback but also that this response was significantly amplified when the participant was responsible for the outcome (playing vs. observing condition).

Finally, we examined the critical interaction between Agency and Painfulness of errors, that is, [Playing-PainfulError > Playing-NonpainfulError] > [Observing-PainfulError > Observing-NonpainfulError]. This yielded four different clusters in the left hemisphere, including AI, but also caudate nucleus, precuneus, and middle frontal gyrus/dlPFC (Table 5). No significant activations were found for the inverse interaction. As illustrated in Figure 4C, the interaction effect in left AI was driven by a large difference between painful versus painless errors in the playing condition, which was absent in the observing condition. Similarly, the left dlPFC was specifically recruited during painful errors in the playing condition (again with no effect of pain during observation).

Table 5. 

Interaction Effect Error Painfulness × Agency, [Playing-PainfulError > Playing-NonpainfulError] > [Observed-PainfulError > Observed-NonpainfulError]

Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Caudate −8 16 286 5.03 
−12 18 4.47 
−2 26 3.35 
AI −26 22 −10 218 4.84 
−32 28 −8 4.55 
Cuneus, precuneus −18 −54 22 361 4.81 
−12 −54 16 4.56 
−28 −64 14 4.19 
Middle frontal gyrus, superior frontal gyrus −16 56 14 389 4.79 
−20 50 18 4.39 
−16 38 26 4.26 
Regions
Left/Right
Peak Voxel Coordinates (mm)
Cluster Size (Voxels)
Peak t Value
x
y
z
Caudate −8 16 286 5.03 
−12 18 4.47 
−2 26 3.35 
AI −26 22 −10 218 4.84 
−32 28 −8 4.55 
Cuneus, precuneus −18 −54 22 361 4.81 
−12 −54 16 4.56 
−28 −64 14 4.19 
Middle frontal gyrus, superior frontal gyrus −16 56 14 389 4.79 
−20 50 18 4.39 
−16 38 26 4.26 

All activations are corrected for multiple comparisons across the whole brain, p < .05 FWE (cluster level).

To exclude the possibility that these interaction effects might just reflect a more general insensitivity of brain responses to observed (compared with self-produced) actions, we also contrasted activations for errors versus correct performance separately for the playing and observing conditions (see Figure 5). This control analysis confirmed that not only the errors caused by the participant (while performing the visual task in the scanner) but also the observed errors (produced by the other participant outside the scanner) activated large and overlapping clusters in areas related to action monitoring, including AI and IFG. Finally, we investigated whether activity in any of the above regions (left AI, dlPFC, dMFC, ventral striatum) was significantly modulated by order effects (i.e., whether participant started the experiment in the scanner and then went outside or vice versa). Mixed-effects ANOVAs on the activity parameters extracted from these regions revealed no significant main effect of Order and no significant interactions between Order and any of the other experimental factors (all p's > .05).

Figure 5. 

SPM results for the main effect of OUTCOME for each condition of agency. Significant activations (p = .05 FEW-corrected at cluster level) are depicted on an inflated brain template for the contrast of “Errors > Correct,” separately for the Playing condition (in red) and the Observation condition (green). Overlapping effects were selectively found in AI and inferior frontal regions.

Figure 5. 

SPM results for the main effect of OUTCOME for each condition of agency. Significant activations (p = .05 FEW-corrected at cluster level) are depicted on an inflated brain template for the contrast of “Errors > Correct,” separately for the Playing condition (in red) and the Observation condition (green). Overlapping effects were selectively found in AI and inferior frontal regions.

DISCUSSION

Previous research, as demonstrated by quantitative voxel-based meta-analysis with NeuroSynth (Yarkoni, Poldrack, Nichols, et al., 2011), has shown consistent overlaps of brain activations during action/error monitoring and pain processing, particularly in dMFC and AI (Shackman et al., 2011; Craig, 2009; Klein et al., 2007). In addition, these regions activate not only to one's own errors or pain but also to observed errors and observed pain in other people, suggesting partly shared neural representations of these events for the self and others (de Bruijn et al., 2009; Singer et al., 2004). However, whether these regions also have a specific role in integrating information across these domains (i.e., self-/other-agency, error monitoring, and social cognition) has remained unclear (Gu, Liu, Van Dam, Hof, & Fan, in press).

Our study sheds light on how these different processes interact during the monitoring of real social behavior. Behavioral results indicated that self-generated painful errors were associated with increased feelings of guilt and higher empathic ratings for pain intensity experienced by the other, two effects demonstrating the significant impact of agency in this experimental situation. Neuroimaging results confirmed the predicted effects of error and pain processing in partly overlapping brain areas, including dMFC and AI. More critically, our data also revealed an important functional dissociation between these regions as a function of agency.

On the one hand, in line with previous studies on error monitoring (Pourtois et al., 2010; Ullsperger et al., 2010; Klein et al., 2007; Taylor et al., 2007), we found selective activations to error feedback in dMFC, but also in AI, plus in the dorsolateral pFC and temporo-occipital areas, whereas the correct outcomes were associated with distinctive increases in ventral striatum. On the other hand, feedback that caused pain to the partner (relative to negative but painless feedback) activated several regions of the “pain matrix” including again both the dMFC and AI, consistent with studies on vicarious pain and empathy (Corradi-Dell'Acqua et al., 2011; Jackson et al., 2005; Singer et al., 2004). Crucially, however, activation in several of these areas was modulated by interactions between outcome, pain, and agency, pointing to an integration of error monitoring with the assessment of affective consequences for others and with self-agency. In particular, agency interacted with signals about performance outcome in dMFC and striatum, but with the vicarious pain feedback in insula and dlPFC. These findings provide novel evidence for a differential contribution of these regions to error monitoring in social contexts.

The ventral striatum is a central part of the dopaminergic mesolimbic system and has a prominent role in the processing of reward and reward prediction errors (Rushworth, Mars, & Summerfield, 2009; Knutson & Cooper, 2005; O'Doherty, 2004; Schultz, 1998). This area typically activates in response to positive outcomes and correct feedback (Vrticka et al., 2008; Ullsperger & von Cramon, 2003) and deactivates in response to negative feedback (Delgado, Nystrom, Fissell, Noll, & Fiez, 2000). Our results showed a differentiation in the ventral striatum between correct and erroneous responses only for the playing condition, suggesting selective tracking of one's own action feedback. No striatum activation was found for correct compared with erroneous observed actions, although monetary rewards (loss or gain of cents) were always presented to both participants as a team. This contrasts with other studies on observation of rewarding outcomes, indicating that striatum responses to vicarious rewards may depend on the social context (de Bruijn et al., 2009; Mobbs et al., 2009). Remarkably, in our study, the ventral striatum did not differ between errors that caused pain versus did not cause pain to the other person. This pattern speaks for a binary self-centered coding of outcomes (correct vs. error) in this region that does not integrate the social consequences of action outcomes.

Another region with agency effects was the dMFC, where activation to errors (vs. correct trials) was enhanced during playing as compared with observing the other person. Considering that negative feedback about one's own action calls for adjustment in behavior (which is not the case when observing performance of another person), these findings converge with recent proposals that the dMFC may play a key role in behavioral control, most crucial for the monitoring of cognitive conflicts (Botvinick et al., 2001) and action outcomes (Ridderinkhof et al., 2004; Rushworth, Walton, Kennerley, & Bannerman, 2004; Holroyd & Coles, 2002). Importantly, in sharp contrast to the striatum, this region was also identified when testing for the main effect of pain. Using meta-analytic evidence, Shackman et al. (2011) recently demonstrated a functional overlap in this region for the processing of pain, negative affect, and response conflict, leading these authors to propose an “Adaptive Control Hypothesis,” according to which dMFC may code for any event requiring behavioral adjustments and control. Our results confirm the overlap in this region for error monitoring and (vicarious) pain processing but also provide new insights on the function of dMFC in action monitoring by showing an increased sensitivity to negative outcomes when these are caused by oneself.

A distinct pattern was found in the left AI. This region responded to both one's own and observed errors, but more importantly also showed an interaction between error agency and error painfulness. A wealth of studies have shown that AI is activated by the observation of others' pain (Corradi-Dell'Acqua et al., 2011; Lamm et al., 2011; Jackson et al., 2005; Singer et al., 2004), as well as by errors and negative feedback (Dhar, Wiersema, & Pourtois, 2011; Ullsperger et al., 2010; Klein et al., 2007). Furthermore, AI is important for the perception of action agency, especially when perceiving the self as a responsible and intentional agent (Brass & Haggard, 2010; Farrer & Frith, 2002). To the best of our knowledge, here we show for the first time that error agency and vicarious pain selectively interact in this region.

These results accord with, but also extend, current theories about pain empathy and theories of cognitive control. In particular, they add to previous suggestions that the representation of another person's pain in AI is not automatic but susceptible to various social and cognitive context factors (Cikara & Fiske, 2011; Hein & Singer, 2008). For example, activation to observed pain is more pronounced for close others and same-group members (Avenanti, Sirigu, & Aglioti, 2010; Cheng, Chen, Lin, Chou, & Decety, 2010; Decety, Echols, & Correll, 2010; Xu, Zuo, Wang, & Han, 2009; Singer et al., 2006), suggesting that interpersonal relationships can moderate the recruitment of shared representations of pain. In particular, our new results challenge the idea that AI responses to vicarious pain are “empathic” in the sense of being automatically triggered by observing pain in another person (Decety & Jackson, 2006; Singer et al., 2004). Indeed, although this region was clearly modulated by observed errors compared with observed correct feedback, no difference in activity was seen between painful and non painful feedback in the observation condition, in contrast to the case of one's own mistakes. Therefore, we surmise that AI responses are primarily driven by the affective relevance for the observer himself: When pain in the other is caused by oneself, it does not only represent an aversive event but also one's own wrongdoing and a violation of social norms. Thus, AI activity might code for the negative emotional impact of incorrect or risky actions (Singer, Critchley, & Preuschoff, 2009; Rilling, King-Casas, & Sanfey, 2008), such as the violation of moral standards and possible threat to social relationships (Chang, Smith, Dufwenberg, & Sanfey, 2011).

Recently, AI sensitivity to errors and negative feedback has been proposed to reflect the conscious perception of negative action outcomes, leading to specific autonomic responses and cognitive adjustment (Dhar et al., 2011; Ullsperger et al., 2010; Klein et al., 2007). Although we did not explicitly assess error awareness, our paradigm (error monitoring based on visual information presented to both participants) ensured that participants were fully aware of the negative (or positive) outcomes in all conditions. Rather than differences in error awareness, we propose that the higher activation of AI for painful compared with painless errors reflects the greater affective relevance and aversiveness of these mistakes (Koban et al., 2012). We do not exclude, however, that awareness might be a prerequisite for such a response or, alternatively, affective aspects could influence error awareness (Ullsperger et al., 2010). The AI receives input from a wide range of cortical areas, including somatic and proprioceptive signals from posterior insula and parietal areas, and plays a major role in the experience of emotions and feeling states (Craig, 2009; Kober et al., 2008) and in the integration of emotion and cognitive effort (Gu et al., 2012). On the basis of these anatomical and functional properties, some authors (Brass & Haggard, 2010) proposed that the AI may integrate information about intentions and decision outcomes to evaluate the affective value of a given action and thus code for “a feeling of what I made happen” (Brass & Haggard, 2010). In keeping with this theoretical account, in our experiment, AI might integrate signals about action agency and error consequences for both the self and other and thus code for the overall affective value of a self-generated action. Such signals may guide subsequent learning and behavioral adaptation (Brass & Haggard, 2010; Lamm & Singer, 2010; Ullsperger et al., 2010) and more generally constitute a cornerstone of moral values and emotions in social contexts.

Finally, a similar modulation was observed in the left dlPFC when one's own errors caused pain to the other. Whereas dMFC is critical for monitoring conflicts and signaling needs for adjustments, the dlPFC seems important for the implementation of cognitive control and post-error behavioral adjustments (Egner & Hirsch, 2005; Cohen, Aston-Jones, & Gilzenrat, 2004; Ridderinkhof et al., 2004; Botvinick et al., 2001). However, our results suggest that these adjustments may not only depend on error signals in dMFC but are also sensitive to the consequences of errors, at least when these are painful for another person. The sensitivity of dMFC to both errors and vicarious pain could induce a greater adjustment in cognitive control and hence interact in dlPFC in this condition. Alternatively, the need for adjustments in top–down control might not only concern errors and performance feedback but also their emotional or social significance. Making mistakes that hurt someone else could generate stronger affective responses and distress, as reflected by higher guilt ratings in the postscanning questionnaire, which might in turn require greater efforts for emotion regulation and attention to refocus on the task demands. Future studies might include trial-by-trial ratings of guilt and emotional arousal in a simplified experimental design, so as to better investigate the affective and social consequences of errors on subsequent cognitive control.

In conclusion, our study goes beyond previous research on error processing in single individuals or interactive social settings by demonstrating that action monitoring systems in the human brain are sensitive not only to the commission or observation of errors but also to their social impact for another person and the potential liability of the agent. Our results reveal a functional dissociation across different regions within brain networks commonly recruited by both action monitoring and vicarious pain. When one's errors cause harm to another person, the dMFC primarily encodes the interaction of agency with negative outcomes; whereas an interaction of error agency with induced pain is selectively observed in the left AI as well as left dlPFC. These integration effects may reflect a greater affective relevance of painful errors, which might call for additional cognitive and affective control. Behaviorally, these mistakes were related to subjective experience of guilt and enhanced pain empathy. More generally, these interactions might provide a neural foundation for affective signals that guide social behavior (Chang et al., 2011).

We note that this novel paradigm allows for several different variations that may be explored in future experiments. For instance, the current study was restricted to the modulation of “empathic” responses to another person's pain as a function of one's actions, but we did not include the reverse condition where errors would cause pain to the person in the scanner. It would also be interesting to investigate how one's own pain is affected by the concurrent processing of self- and/or other-generated errors, as well as to determine how physical punishment could interact with action monitoring and adjustments. Hence, an important avenue for future research will be to better characterize the malleability of nociception by various social and affective factors.

More broadly, these findings illustrate the importance of studying not only isolated cognitive and social processes but also their integration in specific brain systems. By building on—so far—separate research areas of agency, action monitoring, and pain empathy, our novel paradigm opens valuable avenues for investigating social emotions and moral decisions in normal or clinical populations, where empathic processing and action monitoring are disturbed.

Acknowledgments

This research was supported by the National Center of Competence in Research Affective Sciences financed by the Swiss National Science Foundation (51NF40-104897) and hosted by the University of Geneva and a research award from the Evens Foundation. We would like to thank David Sander and Julien Deonna for helpful discussions.

Reprint requests should be sent to Leonie Koban, University of Geneva, Centre Interfacultaire de Sciences Affectives, rue des Battoirs 7, CH-1205 Geneva, Switzerland, or via e-mail: leonie.koban@unige.ch.

REFERENCES

Avenanti
,
A.
,
Sirigu
,
A.
, &
Aglioti
,
S. M.
(
2010
).
Racial bias reduces empathic sensorimotor resonance with other-race pain.
Current Biology
,
20
,
1018
1022
.
Botvinick
,
M. M.
,
Braver
,
T. S.
,
Barch
,
D. M.
,
Carter
,
C. S.
, &
Cohen
,
J. D.
(
2001
).
Conflict monitoring and cognitive control.
Psychological Review
,
108
,
624
652
.
Brass
,
M.
, &
Haggard
,
P.
(
2010
).
The hidden side of intentional action: The role of the anterior insular cortex.
Brain Structure & Function
,
214
,
603
610
.
Chang
,
L. J.
,
Smith
,
A.
,
Dufwenberg
,
M.
, &
Sanfey
,
A. G.
(
2011
).
Triangulating the neural, psychological, and economic bases of guilt aversion.
Neuron
,
70
,
560
572
.
Cheng
,
Y.
,
Chen
,
C.
,
Lin
,
C.-P.
,
Chou
,
K.-H.
, &
Decety
,
J.
(
2010
).
Love hurts: An fMRI study.
Neuroimage
,
51
,
923
929
.
Cikara
,
M.
, &
Fiske
,
S. T.
(
2011
).
Bounded empathy: Neural responses to outgroup targets' (mis)fortunes.
Journal of Cognitive Neuroscience
,
23
,
3791
3803
.
Cohen
,
J. D.
,
Aston-Jones
,
G.
, &
Gilzenrat
,
M. S.
(
2004
).
A systems-level perspective on attention and cognitive control: Guided activation, adaptive gating, conflict monitoring, and exploitation vs. exploration.
In Posner MI (Ed.)
,
Cognitive Neuroscience of Attention
(pp.
71
90
).
New York
:
Guilford Press
.
Corradi-Dell'Acqua
,
C.
,
Hofstetter
,
C.
, &
Vuilleumier
,
P.
(
2011
).
Felt and seen pain evoke the same local patterns of cortical activity in insular and cingulate cortex.
The Journal of Neuroscience
,
31
,
17996
18006
.
Craig
,
A. D. B.
(
2009
).
How do you feel—Now? The anterior insula and human awareness.
Nature Reviews Neuroscience
,
10
,
59
70
.
Critchley
,
H. D.
,
Wiens
,
S.
,
Rotshtein
,
P.
,
Ohman
,
A.
, &
Dolan
,
R. J.
(
2004
).
Neural systems supporting interoceptive awareness.
Nature Neuroscience
,
7
,
189
195
.
de Bruijn
,
E. R. A.
,
de Lange
,
F. P.
,
von Cramon
,
D. Y.
, &
Ullsperger
,
M.
(
2009
).
When errors are rewarding.
The Journal of Neuroscience
,
29
,
12183
12186
.
Decety
,
J.
,
Echols
,
S.
, &
Correll
,
J.
(
2010
).
The blame game: The effect of responsibility and social stigma on empathy for pain.
Journal of Cognitive Neuroscience
,
22
,
985
997
.
Decety
,
J.
, &
Jackson
,
P. L.
(
2006
).
A social-neuroscience perspective on empathy.
Current Directions in Psychological Science
,
15
,
54
58
.
Delgado
,
M. R.
,
Nystrom
,
L. E.
,
Fissell
,
C.
,
Noll
,
D. C.
, &
Fiez
,
J. A.
(
2000
).
Tracking the hemodynamic responses to reward and punishment in the striatum.
Journal of Neurophysiology
,
84
,
3072
3077
.
Dhar
,
M.
,
Wiersema
,
J. R.
, &
Pourtois
,
G.
(
2011
).
Cascade of neural events leading from error commission to subsequent awareness revealed using EEG source imaging.
PLoS One
,
6
,
e19578
.
Egner
,
T.
, &
Hirsch
,
J.
(
2005
).
Cognitive control mechanisms resolve conflict through cortical amplification of task-relevant information.
Nature Neuroscience
,
8
,
1784
1790
.
Farrer
,
C.
, &
Frith
,
C. D.
(
2002
).
Experiencing oneself vs. another person as being the cause of an action: The neural correlates of the experience of agency.
Neuroimage
,
15
,
596
603
.
Fox
,
P. T.
,
Parsons
,
L. M.
, &
Lancaster
,
J. L.
(
1998
).
Beyond the single study: Function/location metanalysis in cognitive neuroimaging.
Current Opinion in Neurobiology
,
8
,
178
187
.
Gu
,
X.
,
Liu
,
X.
,
Van Dam
,
N. T.
,
Hof
,
P. R.
, &
Fan
,
J.
(
in press
).
Cognition-emotion integration in the anterior insular cortex.
Cerebral Cortex
.
Hein
,
G.
, &
Singer
,
T.
(
2008
).
I feel how you feel but not always: The empathic brain and its modulation.
Current Opinion in Neurobiology
,
18
,
153
158
.
Holroyd
,
C. B.
, &
Coles
,
M. G. H.
(
2002
).
The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity.
Psychological Review
,
109
,
679
709
.
Jackson
,
P. L.
,
Meltzoff
,
A. N.
, &
Decety
,
J.
(
2005
).
How do we perceive the pain of others? A window into the neural processes involved in empathy.
Neuroimage
,
24
,
771
779
.
Klein
,
T. A.
,
Endrass
,
T.
,
Kathmann
,
N.
,
Neumann
,
J.
,
von Cramon
,
D. Y.
, &
Ullsperger
,
M.
(
2007
).
Neural correlates of error awareness.
Neuroimage
,
34
,
1774
1781
.
Knutson
,
B.
, &
Cooper
,
J. C.
(
2005
).
Functional magnetic resonance imaging of reward prediction.
Current Opinion in Neurology
,
18
,
411
417
.
Koban
,
L.
,
Pourtois
,
G.
,
Bediou
,
B.
, &
Vuilleumier
,
P.
(
2012
).
Effects of social context and predictive relevance on action outcome monitoring.
Cognitive, Affective & Behavioral Neuroscience
,
12
,
460
478
.
Koban
,
L.
,
Pourtois
,
G.
,
Vocat
,
R.
, &
Vuilleumier
,
P.
(
2010
).
When your errors make me lose or win: Event-related potentials to observed errors of cooperators and competitors.
Social Neuroscience
,
5
,
360
374
.
Kober
,
H.
,
Barrett
,
L. F.
,
Joseph
,
J.
,
Bliss-Moreau
,
E.
,
Lindquist
,
K.
, &
Wager
,
T. D.
(
2008
).
Functional grouping and cortical-subcortical interactions in emotion: A meta-analysis of neuroimaging studies.
Neuroimage
,
42
,
998
1031
.
Lamm
,
C.
,
Decety
,
J.
, &
Singer
,
T.
(
2011
).
Meta-analytic evidence for common and distinct neural networks associated with directly experienced pain and empathy for pain.
Neuroimage
,
54
,
2492
2502
.
Lamm
,
C.
, &
Singer
,
T.
(
2010
).
The role of anterior insular cortex in social emotions.
Brain Structure & Function
,
214
,
579
591
.
Miltner
,
W. H. R.
,
Brauer
,
J.
,
Hecht
,
H.
,
Trippe
,
R.
, &
Coles
,
M. G. H.
(
2004
).
Parallel brain activity for self-generated and observed errors.
In M. Ullsperger & M. Falkenstein (Eds.)
,
Errors, conflicts, and the brain. Current opinions on performance monitoring
(pp.
124
129
).
Leipzig
:
MPI of Cognitive Neuroscience
.
Mobbs
,
D.
,
Yu
,
R.
,
Meyer
,
M.
,
Passamonti
,
L.
,
Seymour
,
B.
,
Calder
,
A. J.
,
et al
(
2009
).
A key role for similarity in vicarious reward.
Science
,
324
,
900
.
Ochsner
,
K. N.
,
Zaki
,
J.
,
Hanelin
,
J.
,
Ludlow
,
D. H.
,
Knierim
,
K.
,
Ramachandran
,
T.
,
et al
(
2008
).
Your pain or mine? Common and distinct neural systems supporting the perception of pain in self and other.
Social Cognitive and Affective Neuroscience
,
3
,
144
160
.
O'Doherty
,
J. P.
(
2004
).
Reward representations and reward-related learning in the human brain: Insights from neuroimaging.
Current Opinion in Neurobiology
,
14
,
769
776
.
Pourtois
,
G.
,
Vocat
,
R.
,
N'Diaye
,
K.
,
Spinelli
,
L.
,
Seeck
,
M.
, &
Vuilleumier
,
P.
(
2010
).
Errors recruit both cognitive and emotional monitoring systems: Simultaneous intracranial recordings in the dorsal anterior cingulate gyrus and amygdala combined with fMRI.
Neuropsychologia
,
48
,
1144
1159
.
Radke
,
S.
,
de Lange
,
F. P.
,
Ullsperger
,
M.
, &
de Bruijn
,
E. R. A.
(
2011
).
Mistakes that affect others: An fMRI study on processing of own errors in a social context.
Experimental Brain Research
,
211
,
405
413
.
Ridderinkhof
,
K. R.
,
Ullsperger
,
M.
,
Crone
,
E. A.
, &
Nieuwenhuis
,
S.
(
2004
).
The role of the medial frontal cortex in cognitive control.
Science
,
306
,
443
447
.
Rilling
,
J. K.
,
King-Casas
,
B.
, &
Sanfey
,
A. G.
(
2008
).
The neurobiology of social decision-making.
Current Opinion in Neurobiology
,
18
,
159
165
.
Rushworth
,
M. F. S.
,
Mars
,
R. B.
, &
Summerfield
,
C.
(
2009
).
General mechanisms for making decisions?
Current Opinion in Neurobiology
,
19
,
75
83
.
Rushworth
,
M. F. S.
,
Walton
,
M. E.
,
Kennerley
,
S. W.
, &
Bannerman
,
D. M.
(
2004
).
Action sets and decisions in the medial frontal cortex.
Trends in Cognitive Sciences
,
8
,
410
417
.
Schultz
,
W.
(
1998
).
Predictive reward signal of dopamine neurons.
Journal of Neurophysiology
,
80
,
1
27
.
Shackman
,
A. J.
,
Salomons
,
T. V.
,
Slagter
,
H. A.
,
Fox
,
A. S.
,
Winter
,
J. J.
, &
Davidson
,
R. J.
(
2011
).
The integration of negative affect, pain and cognitive control in the cingulate cortex.
Nature Reviews Neuroscience
,
12
,
154
167
.
Shane
,
M. S.
,
Stevens
,
M.
,
Harenski
,
C. L.
, &
Kiehl
,
K. A.
(
2008
).
Neural correlates of the processing of another's mistakes: A possible underpinning for social and observational learning.
Neuroimage
,
42
,
450
459
.
Singer
,
T.
,
Critchley
,
H. D.
, &
Preuschoff
,
K.
(
2009
).
A common role of insula in feelings, empathy and uncertainty.
Trends in Cognitive Sciences
,
13
,
334
340
.
Singer
,
T.
,
Seymour
,
B.
,
O'Doherty
,
J.
,
Kaube
,
H.
,
Dolan
,
R. J.
, &
Frith
,
C. D.
(
2004
).
Empathy for pain involves the affective but not sensory components of pain.
Science
,
303
,
1157
1162
.
Singer
,
T.
,
Seymour
,
B.
,
O'Doherty
,
J. P.
,
Stephan
,
K. E.
,
Dolan
,
R. J.
, &
Frith
,
C. D.
(
2006
).
Empathic neural responses are modulated by the perceived fairness of others.
Nature
,
439
,
466
469
.
Taylor
,
S. F.
,
Stern
,
E. R.
, &
Gehring
,
W. J.
(
2007
).
Neural systems for error monitoring: Recent findings and theoretical perspectives.
The Neuroscientist
,
13
,
160
172
.
Ullsperger
,
M.
,
Harsay
,
H. A.
,
Wessel
,
J. R.
, &
Ridderinkhof
,
K. R.
(
2010
).
Conscious perception of errors and its relation to the anterior insula.
Brain Structure & Function
,
214
,
629
643
.
Ullsperger
,
M.
, &
von Cramon
,
D. Y.
(
2003
).
Error monitoring using external feedback: Specific roles of the habenular complex, the reward system, and the cingulate motor area revealed by functional magnetic resonance imaging.
The Journal of Neuroscience
,
23
,
4308
4314
.
van Schie
,
H. T.
,
Mars
,
R. B.
,
Coles
,
M. G. H.
, &
Bekkering
,
H.
(
2004
).
Modulation of activity in medial frontal and motor cortices during error observation.
Nature Neuroscience
,
7
,
549
554
.
Vrticka
,
P.
,
Andersson
,
F.
,
Grandjean
,
D.
,
Sander
,
D.
, &
Vuilleumier
,
P.
(
2008
).
Individual attachment style modulates human amygdala and striatum activation during social appraisal.
PloS ONE
,
3
,
e2868
.
Xu
,
X.
,
Zuo
,
X.
,
Wang
,
X.
, &
Han
,
S.
(
2009
).
Do you feel my pain? Racial group membership modulates empathic neural responses.
The Journal of Neuroscience
,
29
,
8525
8529
.
Yarkoni
,
T.
,
Poldrack
,
R. A.
,
Nichols
,
T. E.
,
Van Essen
,
D. C.
, &
Wager
,
T. D.
(
2011
).
Large-scale automated synthesis of human functional neuroimaging data.
Nature Methods
,
8
,
665
670
.
Yarkoni
,
T.
,
Poldrack
,
R. A.
,
Van Essen
,
D. C.
, &
Wager
,
T. D.
(
2011
).
Cognitive neuroscience 2.0: Building a cumulative science of human brain function.
Trends in Cognitive Sciences
,
14
,
489
496
.