Our study provides several new results. Firstly, using a pseudo-social interaction paradigm, we show that brain responses to facial expressions are strongly modulated by the perceived social meaning
induced by the current context. Smiling faces enhanced activation in the ventral striatum and related regions only when associated with positive feedback, whereas angry faces increased activation in amygdala only when associated with negative feedback. This indicates that responses in both striatum and amygdala were influenced by the social relevance of rewarding and punishment signals expressed by faces, respectively. While many studies have shown activation in striatum and OFC to various types of rewards such as gains or food, a few others have reported activation in the same regions to smiling or attractive faces 
. Here we found that such responses were not driven by facial features alone, but reflected the social meaning of a smiling expression, i.e., when perceived as rewarding current performance and as congruent with task-goals. Similarly, while several studies have reported activation of the amygdala to angry or negative facial expressions, here we show that this response may not be automatic and driven by specific facial features 
but determined by the personal significance of perceived anger.
These findings provide support for the importance of appraisal of personal relevance in emotional processing 
. Moreover, these data also demonstrate that participants were highly motivated by the task (as also confirmed during debriefing). Indeed we observed a reliable main effect of positive (WON) or negative (LOST) feedback in regions associated with reward and motivation processes, including basal ganglia, OFC, and dorsolateral prefrontal cortex for WON trials, as well as retrosplenial cortex and right insula for LOST trials 
. By contrast, we did not find main effects of facial expressions (smiling or angry), but only interactions of expressions with feedback context that confirm that our task induced specific social appraisals as a function of the pseudo-game context. In other words, the brain response to a visually similar facial expression was crucially dependent on current task-goals and feedback congruency, since the social significance of smiles on SF-W trials (praising success) clearly differed from SF-L trials (mocking a failure), whereas anger also differed between AF-L (reproach or punishment) and AF-W trials (conflict or frustration), resulting in distinct patterns of brain responses.
Secondly, and most importantly, we found that individual differences in adult attachment style strongly modulated responses to facial expressions in brain regions associated with affect and motivation; and that such modulation specifically concerned those conditions related to social appraisal. Our results therefore provide new support to the view that adult attachment style can shape how individuals perceive social information in various contexts, and converge with recent behavioral findings that it may modulate recognition judgments for emotional expressions in unfamiliar faces 
. These results also reveal that distinct neural substrates may underlie the two major dimensions of the attachment construct in healthy adults (anxiety and avoidance, as defined by self-report measures used here).
In the congruent feedback condition of social support (SF-W), we found that higher scores on attachment avoidance (AVS) predicted lower activation in brain regions linked to dopaminergic function and reward, including both ventral striatum and VTA 
. This correlation with AVS was highly specific for the SF-W condition representing a socially rewarding interaction (but not related to reward or positive affect of faces alone), supporting the hypothesis that individuals scoring higher on AVS may show reduced activation of affective processes in response to positive social signals 
. These findings are in line with behavioral evidence that people with high AVS tend to prefer physical and emotional distance from others, and usually do not seek social support 
. High AVS is also associated with greater self-reliance and a tendency to dismiss the benefits of group interactions 
. Here we show that such tendencies to avoidant attachment may entail a relative down-regulation of reward-related activity in striatal circuits during socially reinforcing interactions, presumably underlying at least in part the relative impassiveness of individuals with high AVS to social rewards. Our results also provide a plausible substrate for behavioral observations that high AVS is negatively correlated with reward dependence 
, and add support to recent proposals that some forms of social avoidance may be associated with reduced positive experiences in social and non-social contexts 
On the other hand, we found that higher scores on anxious attachment (AXS) were correlated with selective increases in left amygdala responses to social signals of reproach or punishment (i.e. angry expressions combined with congruent negative feedback, AF-L). These data reveal that processing of socially aversive situations is specifically enhanced in brain systems associated with emotional arousal and fear 
for people with higher anxious attachment. Because the amygdala is particularly implicated in processing self-relevant affective information 
, our findings support the notion that a key aspect of anxious attachment may involve enhanced vigilance towards emotionally-significant social cues 
. These condition-specific responses also accord with the view that anxious attachment involves a “relation-specific anxiety” that is distinct from more general forms of anxiety or neuroticism 
. In keeping with this, people with high AXS typically show increased monitoring and exaggerated appraisal of threats to the self, intensify negative emotional responses to emotional or social events, and unlike subjects with high AVS, tend to search more for external sources of support and comfort 
. These results also converge with recent findings that greater amygdala responses to negative sentences may relate to attachment insecurity 
, although the latter study did not examine the distinct prototypes of attachment as here, but inferred more general attachment differences (secure or insecure) based on reaction times to the sentences (slow or fast).
Importantly, note that even though the AF-L condition represented negative social feedback, it was nevertheless congruent with the goals and expectations of participants on LOST trials, and thus did not correspond to a condition of social rejection or exclusion as implemented in other paradigms 
. Here, angry faces were perceived as in-group partners or allies who disapproved failures in the task and hence expressed punishment–a condition meant to activate the need for support in challenging or distressing situations that is intrinsic to anxious attachment style 
. Accordingly, this condition evoked selective activation in the amygdala, rather than in anterior cingulate cortex as reported in previous studies where social rejection implied group exclusion or conflict 
In our study, the third prototype of adult attachment style (secure) did not exhibit any unique correlate for neural responses to the perceived social meaning of facial expressions, but mirrored the pattern found for AVS and AXS, respectively. Thus, high scores on SAS correlated positively with activation of the ventral striatum to rewarding smiles (SF-W) and negatively with activation of the amygdala to reproach faces (AF-L). These data therefore accord with the theoretical view that secure attachment may correspond to a combination of low anxiety and low avoidance, and add new neurobiological evidence in support of bi-dimensional models postulating that these two major components may account for the different categories of adult attachment style 
. Critically, our fMRI results reveal that these two dimensions (anxious and avoidant attachment) have distinct neural bases in two key brain systems implicated in affect and motivation, centered on the amygdala and striatum, respectively. Both the striatum and amygdala play important roles for learning and predicting motivational outcomes in specific situational contexts, and might therefore be well suited for the establishment of idiosyncratic affective responses to social cues based on past experience or developmental history.
Thus, although the exact correspondence between developmental aspects of attachment initially described in infancy 
and attachment style in adults is still partly unclear 
, our results demonstrate that this social psychological construct taps specific affective processes, with distinct neural substrates, which can influence how people automatically perceive and respond to social signals in interaction contexts, beyond relationships with intimate partners or close personal acquaintances 
. In line with our findings, adult attachment has been shown to affect the recognition of emotional expressions in morphs of unfamiliar faces 
, especially when such expressions are relevant to attachment concerns and interpersonal bonding 
, as in our pseudo-social game paradigm. The current imaging findings that brain regions activated by face expressions are differentially modulated by individual attachment style provide new insights on the neurobiological underpinnings of these effects. More generally, unveiling such links between fundamental social dimensions and brain function may not only validate traditional psychosocial conceptualization but also help understand their impact on human behavior.
Activation of STS and MPFC were found only in incongruent feedback conditions corresponding to social opposition or confrontation (AF-W and SF-L), but did not correlate with attachment traits. STS is implicated in theory of mind and perception of intentionality 
, suggesting that participants were more inclined to imagine particular mental states or intentions for faces seen with incongruent feedback information. However, activity in STS did not appear to subtend differences in “mental models” of others that are typically associated with different attachment styles 
. On the other hand, incongruent feedback with angry faces on WIN trials (AF-W) also activated MPFC and vACC, previously implicated in responses to social inclusion-exclusion and emotional conflict 
. Moreover, activity in MPFC and vACC correlated with AVS and overlapped with similar regions activated by social rejection
or emotion suppression 
in other paradigms, suggesting that affective evaluation processes responding to conflict situations might be more active in avoidant subjects in keeping with their more negative appraisal of others 
. This might further contribute to the reduced sensitivity to social reward observed in these subjects.
In sum, our study shows that the two dimensions of adult attachment have distinct neural substrates and produce specific effects on the appraisal of social facial signals. Ventral striatum and VTA were selectively activated by the rewarding feedback value of smiling faces accompanying a success, and thus representing social reward, but this response was blunted in individuals with high AVS scores. Amygdala was selectively activated by the reproach value of angry faces combined with errors, thus representing social punishment, and this response was enhanced in individuals with high AXS scores. In other words, both striatum and amygdala responses were specific to the perceived social meaning of face expressions in relation to current task goals, because no such activity was elicited by the same expressions with a different (incongruent) feedback. Moreover, high AVS also correlated with an increased response to potential social confrontation in ACC, consistent with negative relational schemata hold by avoidant individuals. In contrast, secure attachment was characterized by higher striatal response to rewarding faces and lower amygdala responses to reproach faces, but showed no unique activation pattern, supporting the idea that it may entail a combination of low avoidant and low anxious traits 
. By revealing a critical involvement of emotional brain systems associated with social reward and threat in adult attachment style, our fMRI data provide the first direct neurobiological evidence in support of psychological models proposing two independent affective dimensions to explain these individual differences. More generally, our data also converge with bidimensional models of social disorders that suggest distinct contributions of negative and positive emotions in regulating social behavior and interpersonal communication in a wide range of social contexts 
. Altogether, these results may ultimately help define appropriate intervention strategies in clinical disorders of attachment and social functioning, including autism, phobias, and other relational disturbances.