Social anxiety disorder (SAD), also known as social phobia (SP), is characterized by a severe impairment of social interactions (Ruscio et al., 2008; Harbort et al., 2013). SAD is one of the most common psychiatric disorders, with a lifetime prevalence in the general population of approximately 12% (Alden and Taylor, 2011). It is also known to be frequently accompanied by anxiety and depressive disorders (Beesdo et al., 2007; Delong and Pollack, 2008). There is a wide range of comorbidity rates, depending on the investigated study sample and diagnostic criteria (Harbort et al., 2013).

As fundamental emotional stimuli, emotional faces convey important information in social interactions (Luo et al., 2010; Ran et al., 2014a). An increasing number of electrophysiological studies have investigated the time course of emotional face processing in social anxiety. An early event-related potential (ERP) component, the P100, is assumed to reflect early visual processing and is sensitive to top-down attentional influences (Schendan et al., 1998; Taylor and Khan, 2000). It has been shown that, in individuals with SAD, there is a significant increase in the P100 amplitude in response to threat/angry faces (Mueller et al., 2009). Recently, Peschard et al. (2013) reported that people with SAD demonstrated enhancement of the P100 amplitude for all, not just social, stimuli.

The N170, an occipito-temporal negative deflection, is usually observed in experiments using faces as stimuli (Tortosa et al., 2013; Wiese et al., 2014). There is growing recognition that the N170 reflects a specific attention to the eye region of human faces (Itier et al., 2006, 2007). Numerous ERP studies have shown that faces presented upside down trigger larger N170 amplitudes compared to upright faces (Rossion et al., 1999; Itier and Taylor, 2002). Such amplitude increase with the face inversion is believed to reflect the disruption of early holistic processing stages specific to human faces and has been suggested to be driven by the eye region (Rossion et al., 1999; Itier and Taylor, 2002). In addition to its important role in the face inversion, the eyes are important in conveying different emotions (Itier and Batty, 2009). While some studies failed to detect the moderating effect of SAD on the N170 (Kolassa et al., 2007; Mühlberger et al., 2009), others reported that SAD individuals exhibited more negative N170 amplitudes than controls (Kolassa and Miltner, 2006; Wieser et al., 2010). One can speculate that these inconsistencies may be due to differences in experimental designs and tasks (Peschard et al., 2013).

Subsequent to N170, the P200 component has been suggested to reflect the sustained perceptual processing (Schupp et al., 2004) and the complexity of emotional appraisal (Kolassa et al., 2009). An increasing number of studies have explored how SAD modulates the P200 (Kolassa and Miltner, 2006; van Peer et al., 2010; Gardener et al., 2013). For example, van Peer et al. (2010) reported a specific increase in P200 amplitude to threat faces in SAD individuals, reflecting an attentional bias for social threat. Moreover, a recent study observed a correlation between P200 amplitude in response to self-focus cues and reduced task performance in individuals with SAD (Judah et al., 2015).

As the aforementioned studies demonstrate, the N170 amplitudes were increased in response to threat faces in individuals with SAD compared to controls. This seems to suggest a disruption of holistic face processing in SAD, since the N170 appears to be enhanced when a more feature-based processing is induced (Stahl et al., 2008; Caharel et al., 2011; Ran et al., 2014b). As top-down prediction induces a switch from feature-based to holistic processing (Ran et al., 2014b,c), SAD individuals, with a disturbed face representation, may adopt a holistic coding strategy for perceiving emotional faces when these faces were predictable. This idea is in line with a recent predictive translation hypothesis (Ran et al., 2016b), which proposes that, in individuals with social perception disorders, prior prediction contributes to the normalization of the abnormal processing of social information obtained from faces.

The main goal of the current study was to explore the impact of the top-down prediction on emotional face processing in social anxiety. In accordance with previous literature, we hypothesized that high socially anxious (HSA) participants would exhibit larger N170 amplitudes than low socially anxious (LSA) participants when they perceived unpredictable angry faces. However, when exposed to predictable angry faces, we expected that HSA participants showed no differences in N170 amplitudes compared to LSA participants. In addition, based on the theory of hypervigilance to social threat cues in individuals with SAD (Amir et al., 2009; Klumpp and Amir, 2009), we predicted that there will be increased P100 and P200 amplitudes to angry faces in the HSA compared to LSA participants.

To test these hypotheses, we adopted a variant of the cue-target paradigm that we employed previously (Chen et al., 2015; Ran et al., 2016c). An instruction cue was used to manipulate participants’ prediction bias toward the corresponding emotion. HSA and LSA participants were instructed to perform an emotional task in which angry and happy target faces were presented randomly, and their brain responses were recorded using high temporal resolution ERP techniques.

Investigating the impact of the top-down prediction on emotional face processing in individuals with SAD contributes to the growing body of literature exploring the methods to reduce social anxiety. Moreover, this study could provide a better understanding of the time course of emotional face processing in social anxiety.

The present study employed a modified emotional task in which prediction was manipulated by a cue to investigate how top-down prediction influenced emotional face perception in social anxiety. Behaviourally, we reported an effect of prediction with higher accuracy for predictable than unpredictable faces. Furthermore, we found that HSA but not LSA participants recognized angry faces more accurately than happy faces. For the P100 and P200 components, HSA participants showed enhanced brain activity in response to angry faces compared to happy faces. Moreover, HSA participants exhibited larger N170 amplitudes than LSA participants in the right hemisphere electrodes when they perceived unpredictable, but not predictable, angry faces. The subsequent correlation analysis yielded significant negative correlations between the N170 effect of prediction and the LSAS-SR total scores.

The current work revealed a clear effect of prediction at the behavioral level. Such facilitated response to predictable faces might be due to the fact that prior predictions allow our brains to reduce the number of candidate representations of one object that need to be considered (Kveraga et al., 2007). Moreover, we found that HSA participants were sensitive to angry faces, which is consistent with previous behavioral studies examining social anxiety (Klumpp and Amir, 2009; Stevens et al., 2009; Staugaard, 2010). For instance, Stevens et al. (2009) have found that HSA participants showed an attentional bias toward unambiguously angry faces.

The current electrophysiological results showed larger P100 amplitudes for angry faces in HSA participants when these faces were predictable, suggesting a hypervigilance to angry faces. This finding is consistent with our behavioral results and data from prior ERP studies. Using ERP measures, Mueller et al. (2009) observed that SAD potentiated P100 amplitudes in response to angry-neutral relative to happy-neutral face-pairs. Other studies, however, reported that SAD led to higher P100 amplitudes in response to schematic, artificial and natural facial stimuli, regardless of expression (e.g., Mühlberger et al., 2009; Rossignol et al., 2012; Peschard et al., 2013), suggesting the absence of a specific enhancement to threat in the SAD. While caution should be taken in interpreting these results, one reason for the discrepancy might be that the perception of emotional faces for HSA participants in the present experiment was influenced by the top-down prediction. In the study of Rossignol et al. (2012), participants were instructed to perform a spatial cueing task, while we adopted an emotional cueing task.

Unlike the P100 components, HSA participants exhibited more positive P200 amplitudes in response to angry faces compared with happy faces irrespective of the top-down prediction, which indicated no moderating effect of prediction on the P200 finding of hypervigilance for angry faces. A similar pattern of results has been found in previous ERP studies, which reported hypervigilance toward angry faces for P200 amplitudes in a masked and an unmasked emotional Stroop task (van Peer et al., 2010).

As suggested by Stahl et al. (2008), increased amplitudes and longer latencies of the N170 were associated with disturbed holistic processing of faces. The present study reported that HSA participants exhibited larger N170 amplitudes than LSA participants in the right hemisphere electrodes when they perceived unpredictable angry faces, presumably reflecting a disturbed face representation in HSA participants. The abnormal processing of angry faces was in line with numerous studies examining the moderating effect of social anxiety on the N170 (Kolassa and Miltner, 2006; Wieser et al., 2010). The interpersonal theory of social anxiety (Alden and Taylor, 2004) proposed that SAD is an interpersonal disorder, a condition in which anxiety severely disrupts an individual’s relationships with others. Such tendency for subjects with social anxiety to experience social inhibition may lead to a lack of experience in the perception of social stimuli, such as emotional faces. It is widely accepted that less experience with human faces leads to a relatively more feature-based processing of them (Hugenberg et al., 2007; Short and Mondloch, 2013).

Interestingly, when HSA participants perceived predictable angry faces, they showed no differences in N170 amplitudes compared to LSA participants in the right hemisphere electrodes. This suggests that top-down prediction may improve the deficiencies in building a holistic face representation in HSA participants. Recently, behavioral and neurophysiological studies have reported that the visual stimuli involving mostly feature-based processing were perceived in a holistic manner when these stimuli were predictable (Ran et al., 2014b,c), indicating that prediction induced a switch from feature-based to holistic processing. In this case, HSA participants with a disturbed face representation adopted a holistic coding strategy for perceiving human faces when they were predictable.

There was a significant negative correlation between the N170 effect of prediction and the total scores on the LSAS-SR in both hemisphere electrodes for happy faces in LSA participants. This correlation suggested that LSA participants with more slight anxiety symptoms showed an enhanced ability to predict upcoming positive emotional events. However, no any significant correlation was found in HSA participants. This need to be examined further in the future studies.

Two clusters of interest namely the P100- and N170-cluster were found in the early time window. The P100-cluster had a dipole located in the left occipital gyrus while the N170 cluster showed a generator site in the right occipital gyrus. This suggests that the moderating effect of prediction on the SAD is based on the activation of neuronal populations in the bilateral occipital gyrus. A similar result of source localization has been found in previous fMRI studies, which found that the N170 components were located in the right occipital gyrus (e.g., Qiu et al., 2008). Although some studies have employed the frequency cue to manipulate the prediction (Barbalat et al., 2013; Chen et al., 2015; Ran et al., 2016a,c), such frequency manipulation is not optimal. It is nice to adopt a contingent probability design in further research as it is more informative than the frequency manipulation.

Although a wealth of research has examined facial processing (Luo et al., 2010; Schmitz et al., 2012; Ran et al., 2014a), there is no study that directly investigates the influences of the top-down prediction on the emotional face perception in social anxiety. In the present study, we reported a clear effect of prediction at the behavioral level. The ERP results from the P100 showed larger amplitudes for angry faces in HSA participants when the faces were predictable, suggesting a hypervigilance to angry faces. Unlike the results in P100, HSA participants exhibited more positive P200 amplitudes for angry compared with happy faces irrespective of the top-down prediction, indicating no moderating effect of prediction on the P200 finding of hypervigilance for angry faces. Crucially, HSA participants exhibited larger N170 amplitudes than LSA participants in the right hemisphere electrodes when they perceived unpredictable, but not predictable, angry faces. This result suggests that top-down prediction may improve the deficiency in building a holistic face representation in HSA participants.

GR and XC designed the experiments. GR collected and analyzed the data. GR primarily wrote the manuscript. All authors discussed the results and commented on the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.