Social anxiety disorder (SAD) is defined by a marked and persistent fear of social or performance situations in which the person is exposed to unfamiliar people, or possible scrutiny by others [1]. When interacting with others, people with SAD experience intense anxiety accompanied by physical symptoms such as blushing, heart racing, sweating, trembling, and nausea, as well as stuttering and trouble concentrating [2]. SAD is one of the most common anxiety disorders, with a lifetime prevalence of up to 13% [3]. When untreated, SAD can lead to elevated social isolation, depression, anxiety, and alcohol and/or substance abuse [4].

Models of SAD have highlighted the role of emotional hyper-reactivity in feared social situations. Exaggerated emotional responses are thought to arise from maladaptive appraisals of the social situation that transform innocuous social cues into interpersonal threats [5]. This hyper-reactivity often leads to attempts to escape from or avoid the anxiety-provoking object or situation [2].

To study the behavioral and brain aspects of this emotional hyper-reactivity, most studies have employed negatively valenced facial expressions as socio-emotional probes. Behaviorally, these studies show hyper-arousal and increased emotional distress in patients with SAD when exposed to negative face stimuli [2]. Neurally, however, the findings are less clear.

The amygdala, which plays a critical role in the recognition of fear in facial expressions [6], was a natural target in studies of the neural basis of emotional reactivity in SAD. It was suggested that the heightened emotional responses in SAD patients were associated with increased amygdala activity [7]. Amaral [8] proposed a model based on lesion studies in monkeys, which suggested that the amygdala performs a protective function, namely, allowing the organism to detect and avoid danger. This model provided a rationale for theories implicating the amygdala in the neuropathology of SAD.

Many studies have indeed demonstrated heightened amygdala responses in patients with SAD when viewing angry, fearful and even neutral facial expressions [7, 9–12]. However, other studies, some using stimuli other than faces, failed to detect differential amygdala activation between patients with SAD and HC [13–17]. For example, Quadflieg and colleagues (2008) have found that both SAD patients and HC recruit the amygdala when listening to angry compared to neutral prosody [17]. Similarly, Goldin and colleagues (2009) found increased amygdala responses to harsh facial expressions with longer stimulus presentation times (6 seconds) in both SAD patients and healthy controls [16]. Other studies have even found decreased amygdala responses in SAD patients compared to HC. For example, Kilts and colleagues (2006) have found decreased amygdala responses during both script-driven imagery of anxiety-provoking social situations and a mental arithmetic task in SAD [14].

A second limbic region that is thought to play a prominent role in SAD neuropathology is the insula. The insula is interconnected with the amygdala, thalamus, and orbitofrontal cortex [18], regulates the autonomic nervous system [19], and is involved in the recognition and experience of aversive states [20] and threat processing [21]. Thus, hyper-reactivity of the insula in SAD patients might be responsible for the generation of fear responses to social aversive stimuli.

A recent meta-analysis identified four functionally distinct sub-regions in the human insula implicated in social-emotional, sensorimotor, olfacto-gustatory, and cognitive brain networks [22]. The ventral part of the insula was associated with the social-emotional domain. This ventral region was active in 195 experiments of emotion generation (paradigms of “induction, imagination or recall of own happiness, fear, anxiety, anger, sadness, or disgust”) and 120 empathy related experiments (paradigms of judging emotions in faces or attending to pain in others). This differentiation matches previous cytoarchitectonic studies in non-human primates that found differences between the anterior-basal agranular, allocortical, and posterior granular parts of the insula [23]. The anterior-ventral region is involved in the generation and mediation of feeling as a response to environmental stimuli and affective states [24], and in the recognition of emotions in faces [25]. In SAD, increased activity in this insula region was associated with visualization of angry and disgusted faces [13, 26], viewing of images with negative emotional content [27], and anticipation of a public speaking task [28]. However, several studies did not find such activity differences [17, 29] and others found decreased activity in the insula during a public speaking task [30] and during an implicit learning test [31].

Thus, despite the clear focus of neuroimaging studies on the amygdala and insula as key regions in SAD pathology, findings are mixed. One explanation for these mixed findings might be related to factors such as different stimuli and experimental designs used to elicit emotional responses (mainly emotional faces until a few years ago, and images, written words, and auditory stimuli more recently), small patients groups, differences in exclusion criteria (e.g., different psychiatric co-morbidity), and other experimental parameters.

However, another explanation could be a lack of concordance between the behavioral and physiological components of anxiety. This idea was raised by Mauss and colleagues in a study of non-clinical social anxiety [32]. In this study, anxiety experience, subjectively perceived physiological activation, and objectively measured physiological activation were assessed in high and low trait social anxiety participants using an impromptu speech paradigm. The study found no differences in physiological responding between high and low trait social anxiety, despite clear differences between the two groups in their subjective experience of anxiety and their perceived physiological activation. Although this study did not measure brain responses, and did not employ a clinical sample, it did raise an important question about the widely assumed link between behavioral and physiological responses in the context of social anxiety.

In light of the mixed literature on the neural correlates of emotion reactivity in SAD, the goal of the present study was to examine both behavioral responses and BOLD signal changes in SAD compared to HC when reacting to distinct types of socio-emotional stimuli. To test whether behavior and brain responses vary by context, we used three different socio-emotional tasks: 1) viewing looming harsh faces (Faces), 2) watching dynamic videos of actors delivering social criticism (Criticism), and 3) reading negative self-beliefs (Beliefs). These tasks mirror the specific types of interpersonal evaluations and negative thoughts that are especially salient in patients with SAD. As suggested by the literature, we focused on the amygdala and the insula regions, but also performed whole-brain analyses to examine the involvement of other brain networks in SAD.

We hypothesized that, compared to HC, patients with SAD would have (1) greater self-reported negative emotion reactivity for each of the three tasks; and (2) greater BOLD responses in the amygdala and insula. In secondary analyses, we examined the differential effect of each socio-emotional task on negative emotion reactivity and neural responses in SAD and HC, and whether – among SAD participants – negative emotion reactivity and neural responses were associated with social anxiety symptom severity.

The goal of the present study was to investigate behavioral and brain responses in patients with SAD versus HC when reacting to three socio-emotional tasks (faces, criticism, beliefs). As we expected, at the behavioral level all three tasks induced heightened negative reactivity in SAD versus HC participants. However, we found no evidence of between-group differences in amygdala and insula BOLD responses. Both SAD and HC had increased left and right amygdala responses in all three tasks, and increased left and right insula responses in 2 of the 3 tasks (Criticism and Beliefs).

How might this discrepancy between our current results and previous work on emotion-related brain responses in SAD be explained? One possibility is that differences in the experimental tasks used played a crucial role [43]. To date, most studies have used static facial expression stimuli to examine neural processing of interpersonal threat cues. Studies that have investigated social stimuli other than faces have yielded mixed results [16, 17, 44, 45]. For example, no differential recruitment of the amygdala and insula in SAD and HC were found when listening to angry versus neutral voices [17], when performing a social situation task [44], or when playing a ‘decision making’ game with others [45].

In our study, we used three distinct types of socio-emotional tasks with relatively high levels of complexity. The Faces task used a static emotional facial expression with the addition of looming (i.e., the appearance of approaching the perceiver). The Criticism task used dynamic videos of an actor delivering socially critical comments. The Beliefs task used participant-generated negative self-beliefs embedded in an autobiographical script about a socially painful interpersonal situation. In addition, all three tasks paradigms had longer stimulus presentation durations (9–12 sec) than have been used in most prior studies.

It is possible that the increased stimulus complexity and presentation durations in our study increased the likelihood that healthy subjects would evaluate the stimuli as threatening, and thus may have yielded emotion reactivity related brain responses similar to those generated by the patients with SAD. In addition to the stimuli used to elicit emotional responses, other design parameters such as whether the participants are instructed to simply look at the stimuli, or whether they have to perform a task, and whether the task is implicit or explicit, may have contributed to the observed findings [46].

A second possible explanation of our findings builds upon the first, and suggests that at least in the complex socio-emotional contexts we employed in our study (and which characterize many everyday life contexts), increased negative emotion ratings in SAD may be related to exaggerated cognitive biases, including negative self-reflective and ruminative processes. It is also possible that there are differential tendencies to implement spontaneous (uninstructed) emotion regulation directed at the emotion experience in HC versus SAD patients. This could have yielded lower levels of negative emotion ratings in HC. Thus, if the normative pattern shown in HC is to engage in automatic emotion regulation, then one feature of SAD is the absence of such un-cued, automatic activation of emotion regulation. Therefore, maladaptive cognitive processes, such as rumination, suppression, and self-criticism, and ineffective emotion regulation, in SAD may lead to exaggerated negative emotion experience.

If such accounts are correct, we might expect other brain regions (in addition to amygdala and insula) to show differential response patterns between SAD and HC when exposed to socio-emotional stimuli. Indeed, whole brain analyses revealed SAD versus HC differences in several brain regions, specifically, increased frontal, occipital, and temporal cortical activity in SAD versus HC during the Faces and Criticism tasks, and decreased frontal activity in SAD versus HC during the Beliefs task (Table 3). These regions have been shown to be involved in emotion regulation (middle frontal gyrus) [47], recognition of faces (middle temporal gyrus) [48], language processing (superior temporal gyrus), memory (lingual gyrus, parahippocampus), and social cognition (medial frontal gyrus, superior temporal gyrus) [49]. In SAD, though not much discussed, abnormal activity in these regions has been related to difficulties in reasoning [28], mentalizing abilities [45], and impaired perception of self and others [50–52]. Additionally, the whole brain analyses further support our findings of no between-group differences in amygdala and insula activity when reacting to the socio-emotional tasks.

Congruent with these findings, in a recent study, Doehrmann and colleagues (2012) examined the association between responses to cognitive-behavioral therapy for SAD and pre-treatment brain activations to social (facial) and nonsocial (scene) stimuli [53]. They found pre-treatment activity in visual cortical regions correlated with treatment outcome, and no association with treatment outcome in the amygdala, despite its robust activation to all experimental conditions. The authors explained these results by emphasizing dysfunctional emotion regulation, and specifically attentional deployment, in SAD. They suggested that CBT is perhaps particularly successful in patients with better emotion regulation capacities, which is correlated with already stronger responses to angry faces in visual regions. Thus, in addition to showing no abnormality in baseline amygdala response in SAD, this study suggests that brain regions more related to emotion regulation and not to emotion reactivity, might be the core deficit in SAD, and the focus of change during treatment.

More generally, our finding of increased negativity in SAD compared to HC when facing emotional stimuli, with no such SAD versus HC differences in the amygdala and insula responses, could suggest that self-report of negative emotion is less tightly coupled with increased limbic activity than is typically thought, at least in the context of social anxiety. This hypothesis accords with the findings of Mauss and colleagues [32], namely, a decoupling of subjective emotion and physiological measures and objective physiological responses in high and low socially anxious participants when giving an impromptu speech. The authors suggested that the findings of increased physiological activation during the speech in all participants (and not only in participants who reported increased experience of anxiety) leaves open the possibility that physiological activation might be necessary for the experience of anxiety, but such activation is clearly not sufficient to explain inter-individual variation in anxiety experience. The results of the current study strengthen the conclusions of Mauss and her colleagues. It might be that the relation between subjective feeling and physiological response is more complex than a simple ‘amygdala = emotion’ equation would suggest.

It also bears comment that in the current study, examination of between task effects enabled us to examine whether the behavioral and neural results were related to a particular type of socio-emotional stimuli. Although all three tasks were successful in terms of inducing higher emotional responses in SAD, compared to HC, it seems that no task activated all emotion generation brain regions consistently. Our results indicate that different contexts, i.e. different “anxiety-inducing” experimental probes may have quite different effects, behaviorally and neurally. Behaviorally, the Beliefs task yielded the highest negative emotion reactions in both SAD and HC. Thus, the idiographic, participant-generated negative self-beliefs were the most potent emotional probe, most likely because of their self-relevance or personal salience. These probes were also the most ecologically valid type of stimuli in our study, as negative self-focused automatic thoughts are usually the focus of most forms of psychological interventions for SAD. Interestingly, the Beliefs task, which elicited the greatest negative emotion responses, did not elicit the greatest brain responses. Specifically, right amygdala responses were equally high for Criticism and Faces, and lower for Beliefs (with no differential task effects in the left amygdala) in SAD patients. BOLD responses in the left and right insula were highest when reacting to Criticism > Beliefs > Faces in both SAD and HC.

Contrary to our hypothesis, social anxiety symptom severity was not correlated with negative emotion ratings, and was associated only with greater left insula activation during the Faces task, with no association with amygdala response. This finding accords with the findings of Schmidt and colleagues [46] of a positive correlation between insula activity and symptom severity in SAD patients, despite no differential effect in the insula between SAD and HC, during explicit processing of verbal threat-related stimuli. The association with symptom severity suggests that within patients with SAD, symptom severity matters. Thus, during the Faces task, even though as a group SAD patients had insula responses comparable to the HC group, greater social anxiety symptom severity within patients was found to be associated with increased neural responses. A recent study by Furmark and colleagues further supports the importance of looking beyond only meeting diagnostic criteria for SAD. They found that both patients with SAD and HC had increased left amygdala activation in response to angry compared with neutral faces, but that genotype (serotonergic polymorphisms) and not diagnosis explained a significant portion of the variance in amygdala responsiveness [54].

This is the first study to specifically assess three distinct types of socio-emotional experimental tasks in a large sample of patients with SAD at a single time point. Our main finding is similar increased amygdala and insula responses when reacting to socio-emotional stimuli, despite increased emotional responses in SAD, compared to HC. The fact that we obtained the same result using three different tasks provides strong validation of our findings. To test our hypothesis of deficits in emotion regulation processes in SAD, future studies should investigate regulation processes, both explicit (when instructed to regulate) and implicit (spontaneous), in SAD and in HC. A direct examination of emotion regulation processes and their association with neural activity in limbic and cortical brain regions could help in understanding the deficient appraisal processes in SAD.

Methodologically, more studies are needed to further examine different kinds of tasks/contexts, which might help to evaluate the influence of different variables, such as the degree of personal salience, the intensity of the emotional stimuli, and the modality of the stimuli, on the behavioral and neural responses. Also, in addition to examination of negative emotional stimuli, an interesting venue could be the investigation of positive emotional stimuli, such as social praise, or positive self-related thoughts.

In this study, in order to be able to compare between the different tasks, we contrasted the React condition to an Asterisks condition. However, in the future, it will be important to examine the results when contrasting reactivity to negative stimuli with reactivity to neutral stimuli as well.

Finally, an interesting question arising from the current results is whether other clinical populations such as major depression, generalized anxiety, or mixed anxiety-depression, show the same dissociation between heightened emotional reactivity and similar limbic responses, compared to HC. In the future, it will be important to compare behavioral and BOLD responses during negative emotional contexts in SAD group versus other clinical populations, in order to examine the specificity of the results reported here.