Reduced early neural processing of faces in children and adolescents with social anxiety disorder

Social anxiety disorder (SAD) is one of the most common mental disorders during childhood and adolescence. Yet, little is known about its maintenance in youth. Cognitive models of SAD indicate that attentional biases play a key role in the dysfunctional processing of social information, such as emotional faces. However, previous research investigating neural correlates of childhood SAD has produced inconsistent findings. The current study aims to investigate neural face processing in children and adolescents with SAD, while taking into consideration methodological limitations of previous studies. We measured event-related potentials (P100, N170, EPN, LPP) in response to happy, neutral, and angry adult faces


Introduction
Social anxiety disorder (SAD) is one of the most common mental disorders among children and adolescents (Kessler et al., 2012).It is characterized by an intense fear of negative evaluation by others and often leads to a persistent avoidance of social situations, such as social interactions and performance situations (American Psychiatric Association, 2013).In most cases SAD develops during late childhood and adolescence and is associated with substantial negative psychosocial consequences and high comorbidity rates (Beesdo-Baum & Knappe, 2012;Kessler et al., 2005;Stein & Stein, 2008;Szafranski et al., 2014).Despite the high prevalence of SAD in youth, relatively little is known about the factors involved in the development and maintenance of the disorder during childhood and adolescence (Halldorsson & Creswell, 2017).
Cognitive models of SAD indicate that social anxiety is maintained by a broad range of dysfunctional cognitive processes.These include negative self-beliefs, self-focused attention, and attentional biases before, during, and after social stress (e.g.Clark & Wells, 1995;Rapee & Heimberg, 1997).Regarding attentional processing of social stimuli, such as faces, models assume that socially anxious individuals automatically allocate their early attentional resources toward potentially threatening social information (e.g., facial expressions of negative evaluation) and may show difficulties in disengaging from such information (e.g., Rapee & Heimberg, 1997).This in turn can increase social anxiety and reinforce negative schemata about the social surrounding.Importantly, socially anxious individuals may also respond with avoidance, thereby failing to habituate to anxiety-provoking situations (Heinrichs & Hofmann, 2001).

Face processing in adults with social anxiety
A large body of behavioral studies has demonstrated altered attentional processing of social information, such as angry faces, in adults with high social anxiety (for a review see Bar-Haim et al., 2007;Staugaard, 2010).In addition to behavioral measures, the assessment of event related potentials (ERPs) via EEG can give insights into scalp-recorded neural activity in response to specific stimuli and may shed light on the specific time course of very early and automatic attentional processes.Early, more automatic stages of attentional processing have already been investigated, for example, with the P100.This component reflects an occipital positive deflection that occurs 80-120 ms after stimulus presentation and is indicative of early attention allocation (Luck, 2005).The occipito-temporal N170 represents a negative potential, which is assumed to be involved in the structural pre-categorical encoding of faces and shows face-sensitivity in adults and children (Bentin et al., 1996;Kuefner et al., 2009;Rossion & Jacques, 2008).The early posterior negativity (EPN) is often measured using temporo-occipital electrodes and is modulated by the emotional salience of stimuli (Junghöfer et al., 2017;Schupp et al., 2004).At later stages, beginning around 350 ms after stimulus onset, more conscious and controlled neural processing of faces is frequently assessed by the late positive potential (LPP; Schindler & Bublatzky, 2020;Schupp et al., 2000).The occipito-parietal LPP is assumed to reflect motivated attention, emotion regulation, and sensitivity to the perceived salience of stimuli (Cuthbert et al., 2000;Dennis & Hajcak, 2009;Schupp et al., 2000).
Adult ERP studies on samples with clinical SAD or high levels of social anxiety have yielded heterogenous results (for a review see Harrewijn et al., 2017).The most consistent finding seems to be an association between social anxiety and higher P100 amplitudes in response to social stimuli (Kolassa et al., 2007;Mueller et al., 2009;Mühlberger et al., 2009;Peschard et al., 2013;Rossignol et al., 2012).Findings regarding the N170 are more conflicting.Most studies did not find an association with social anxiety (Kolassa et al., 2007;Mühlberger et al., 2009;Peschard et al., 2013;Rossignol et al., 2012;Schmitz et al., 2012).However, some studies found either lower or higher N170 amplitudes in the social anxiety group (Kolassa & Miltner, 2006;Mueller et al., 2009;Wieser & Moscovitch, 2015).Further, Schmitz et al. (2012) reported enhanced EPN amplitudes in highly socially anxious participants.At later stages of processing, higher LPP amplitudes in response to threatening faces and averted gaze have been found in highly socially anxious samples (Kolassa et al., 2007;Moser et al., 2008;Mühlberger et al., 2009;Schmitz et al., 2012).

Face processing in children and adolescents with SAD
Similar to the findings in adults, current research suggests that anxious youth also show a bias toward threatening stimuli in behavioral paradigms (Dudeney et al., 2015).Additionally, altered visual processing of social information has been observed in 10-to 13-year-old children with SAD (Keil, Hepach, et al., 2018).Nevertheless, it is important to note that findings from adult studies cannot be simply generalized to younger populations due to significant developmental differences.On a behavioral level, threat biases in child samples appear to be less robust and change with age (Dudeney et al., 2015).Relatedly, important brain maturation processes take place during childhood and adolescence that are thought to lead to greater efficiency in information processing and an involvement of different cortical and subcortical structures with age (Luna et al., 2004;Monk, 2008;Yurgelun-Todd & Killgore, 2006).Furthermore, developmental studies on face processing often report age effects, with younger children typically showing elevated ERP amplitudes (Kuefner et al., 2009;Kujawa et al., 2012;MacNamara et al., 2016).
Relatively little research has focused on neural correlates of emotional face processing in children and adolescents with SAD.Schwab & Schienle, (2017, 2018) conducted two important ERP studies to examine the neural response to schematic faces and the impact of social context.In both studies, children with SAD and healthy controls (HCs) did not differ in their early neural processing of faces (P100 and N170) but showed higher LPP amplitudes in response to schematic angry faces, and photos of happy and angry faces, regardless of the social context.Similarly, Kujawa and colleagues (2015) found higher LPP amplitudes in response to angry and fearful faces in 7-to 19-year-olds with mixed anxiety disorders (including SAD) compared to HCs.Enhanced LPP amplitudes were particularly prominent in youth with SAD.However, in a different study (Keil, Uusberg, et al., 2018), no differences among children (aged 10-13 years) with SAD, clinical controls with mixed anxiety disorders, and HCs were found in early and late neural responses (P100, N170, EPN, and LPP) to happy, neutral, and angry faces.Interestingly, exploratory analyses revealed a positive association between age and N170 amplitudes in SAD but not in HC children.This suggests that the neural processing bias in children with SAD increases with age.In another study that included children aged 8 to 12 years with high and low social anxiety, no group effect was found on the P100, but socially anxious children showed lower N170 amplitudes than non-socially anxious children in response to all displayed emotions (Wauthia et al., 2023).

Limitations of previous research on SAD in children and adolescents
Although the abovementioned studies provide important insights into neural face processing in youth with SAD, it is important to note several limitations.First, most of the studies were based on rather small sample sizes (e.g., only 15 to 22 participants per group; Schwab & Schienle, 2017, 2018;Wauthia et al., 2023), which limits the statistical power to detect moderate or small group differences and hinders the assessment of important moderating variables, such as age.Second, several studies investigated samples of children and adolescents with various anxiety disorders, thus limiting conclusions for SAD and possibly masking group differences due to potential differences in anxiety-specific processes (e.g., Kujawa et al., 2015).Third, most studies did not include a clinical control group, which allows no conclusion about transdiagnostic or disorder-specific effects of neural responses to emotional faces.The universal research domain criteria approach proposes that anxiety disorders may not be as distinct as suggested by diagnostic manuals because they share certain biological and cognitive processes (Insel et al., 2010).Accordingly, the inclusion of a homogenous clinical control group would allow for comparisons among specific anxiety disorders.Fourth, some of the study samples were not only small but also comprised a wide age range of both children and adolescents (Kujawa et al., 2015;Schwab & Schienle, 2017, 2018;Wauthia et al., 2023), which could mask potential age-related changes in emotional face processing, as reported by others (Kuefner et al., 2009;Kujawa et al., 2012;MacNamara et al., 2016).Last, to our knowledge, no study has included a control condition of non-social stimuli.This would give insights into the question of whether attentional biases are specific to social stimuli.

The current study
Taking the limitations of previous research into account, in the present study we investigated neural correlates of emotional face processing in children and adolescents with SAD on both behavioral and neural levels.We further considered age effects in a large sample of youth with SAD and compared them both to HCs and a homogenous clinical control group with specific phobias (SP).Our four hypotheses were as follows: (1) We expected youth with SAD to show differences in early attentional processing of emotional faces (P100, N170, and EPN) compared to youth with SP and HCs.In light of inconsistencies in previous findings, we made no assumptions about the direction of the effects.(2) For later stages of attentional processing, we expected higher LPP amplitudes in the SAD group in response to emotional faces compared to youth with SP and HCs (Kujawa et al., 2015;Schwab & Schienle, 2017, 2018).(3) Further, we wanted to explore whether potential group effects were moderated by age.(4) We expected that group differences between youth with SAD and both control groups would be specific to faces and not evident in response to non-social household objects.

Participants
We recruited participants in two German cities via local schools and the cities' register of residents for a larger 2-center project on childhood SAD.The final sample consisted of 159 children and adolescents (82 female) aged 10 to 15 years, in the following referred to as youth.An additional seven children were excluded from final analyses due to incomplete EEG experiments (n = 3), irregularities in the EEG data (n = 1), or based on exclusion criteria (n = 3).The study was approved by the local ethics committee and all youth and their parents were informed about the three-session project procedure and provided informed consent.Children and adolescents received age-appropriate vouchers worth 70€ for their participation, and parents were paid 30€.The inclusion criterion was a primary diagnosis of SAD (SAD group; n = 57) according to DSM-5 criteria, a primary SP diagnosis without comorbid anxiety or mood disorders (SP group; n = 41), or no lifetime mental disorder (HC group; n = 61).Exclusion criteria for all groups included a current or past psychotic episode, severe depressive episodes, suicidality, current or past psychotherapeutic treatment, psychotropic drug intake, pervasive developmental or neurological disorders, strong visual impairment, and an IQ below 80 based on the short version of the culture fair intelligence test (CFT 20-R; Weiß, 2006).Parents of potential, interested participants first completed a brief phone screening.Eligible youth and their parents were then interviewed separately using the Structured Diagnostic Interview for Mental Disorders in Children (Kinder-DIPS; Schneider et al., 2017) and completed several questionnaires to assess their diagnostic status.Trained advanced graduate or doctoral students conducted the diagnostic interviews, which were videotaped and supervised by licensed clinical psychologists.A previous version of the Kinder-DIPS showed very good interrater reliability for lifetime anxiety disorders (parents: k = .94,children: k = .90;Neuschwander et al., 2013).

Psychometric measures
The German version of the Social Anxiety Scale for Children-Revised (SASC-R-D; Melfsen & Florin, 1997) assesses two dimensions of social anxiety via self-report: fear of negative evaluation (SASC-FNE) and social avoidance and distress (SASC-SAD).It consists of 18 items which are rated on a 5-point Likert scale, from 1 (not at all) to 5 (all the time).The scale showed good internal consistency in the current sample (SASC-FNE: α = .87,SASC-SAD: α = .82,total score: α = .92).We measured phobic fears with the German version of the Fear Survey Schedule for Children -Revised (FSSC-R; Döpfner et al., 2006).The self-report consists of 96 items on various fear-related situations and stimuli that are assessed on a three-point scale (0 = none, 1 = some, 2 = a lot).Internal consistency was excellent in the current study (α = .96).
The German version of the Children's Depression Inventory (CDI; Stiensmeier-Pelster et al., 2014) assesses depressive symptoms and their severity in children and adolescents based on 29 self-reported items according to DSM-5 criteria.In the current sample, the CDI showed high internal consistency (α = .91).The Child Behavior Checklist (CBCL; Döpfner et al., 2014) measures emotional and behavioral problems, and somatic complaints in children and adolescents from the age of 6 to 18 years based on parent reporting.The checklist consists of 100 items that can be grouped into internalizing and externalizing symptoms.Internal consistency in the current sample was excellent (internalizing subscale: α = .92,externalizing subscale: α = .83,total score: α = .94).

Stimulus material and procedure
During the EEG recording, participants sat on a comfortable chair in a sound-attenuated, temperature-controlled room and were instructed to watch the images attentively.The experiment consisted of 243 trials arranged into 3 blocks with short breaks in between.We ran four practice trials before the first block started.Overall, 198 adult faces (66 trials for each condition (angry, neutral, happy), 50 % female) from the Karolinska Directed Emotional Faces database (Lundqvist et al., 1998) and 45 pictures of non-social household objects from the Bank of Standardized Stimuli (Brodeur et al., 2014) were presented.Each trial started with a variable inter-trial interval (800-1200 ms) with a fixation cross presented for 500 ms in the center of the screen.Next, the stimuli were shown for 2000 ms.After the presentation of a face, children had to identify the depicted facial expression by choosing between two emojis, one representing the shown emotion and one showing one of the other two emotions in randomized order.When presented with objects, children had to decide whether they saw a face or an object, the choice being between an emoji illustrating a light bulb versus one of the facial expressions.Decisions were carried out by pressing a corresponding key with the left or right index finger while reaction times (RTs) and accuracy were measured.Fig. 1 shows an exemplary trial sequence.The order of the stimuli was randomized with the following restrictions: The same emotion was not shown in more than three consecutive trials and the same face was not presented in two consecutive trials.In the beginning and at the end of the experiment, the children had to rate their state arousal level on an 11-point scale (0 = no anxiety to 10 = extreme anxiety).The stimulus presentation was programmed in Matlab Version 9.7.0 (The MathWorks Inc., 2019), using the Psychophysics Toolbox extensions (Kleiner et al., 2007).

EEG recording and data reduction
Continuous EEG was recorded with a 64-channel active electrode acti-CAP system (actiCHamp Plus, Brain Products GmbH, Gilching, Germany) and Brainvision Recorder at a sampling rate of 1000 Hz.In one study center the BrainAmp-amplifier with a reference electrode located at FCz was used; in the second study center the set-up included a QuickAmp-amplifier with an online averaged reference.In both study centers, electrodes were placed following the 10-20 system with the ground electrode positioned at FPz.In addition, four electrooculogram electrodes were positioned below and above the left eye and at the outer epicanthus of each eye to record horizontal and vertical eye movement.Electrode impedances were kept below 20 kΩ.Offline data was preprocessed using BrainVision Analyzer (BrainVision Analyzer, Version 2.2.0, Brain Products GmbH, Gilching, Germany).After a visual inspection of the data, problematic channels were removed and the remaining channels were re-referenced to an average reference.Next a band-pass filter from 0.1 to 30 Hz was applied.Ocular artifacts were corrected based on an implemented semiautomatic independent component analysis (ICA), which used a restricted infomax algorithm that was trained on artifact free data in an interval of 360 s.After the ICA, removed channels were re-interpolated.EEG data was time-locked to the stimulus onset, segmented into epochs 200 ms pre-stimulus onset to 2000 ms post stimulus onset and a baseline correction was applied.Artefactual epochs were identified based on a threshold criterion of ± 150 μV and a maximal allowed voltage step of 50 μV/ms and then removed.This resulted in an epoch retention rate of 94.92 % (SD = 5.55 %, range 53.33 to 100 %), which did not differ among the groups or conditions (robust ANOVA: ps > .35).
For the ERP components, we chose cortical positions and time windows according to previous studies, visual inspections of scalp distributions (for an overview see Fig. 2), and grand averages.We then extracted the averaged activity within the chosen time windows.We measured P100 at O1 and O2 (Keil et al., 2022;Keil et al., 2018;Kolassa & Miltner, 2006;Schmitz et al., 2012), at 70-140 ms post stimulus onset.We quantified the N170 component between 140 and 190 ms at P7 and P8 (Keil et al., 2022;Keil et al., 2018;Kolassa & Miltner, 2006;Schmitz et al., 2012).For the EPN, we also based the electrode selection on the difference wave between emotional (angry and happy) and neutral faces.Accordingly, the EPN was extracted at POz, PO3 and PO4 in a time widow between 220 and 280 ms.The LPP reflected the mean activity at occipital-parietal sites, namely POz, PO3, PO4, Oz, O1, and O2 and was divided into two time-windows: Early LPP (350-600 ms) and late LPP (600-1000 ms).Finally, we averaged epochs across electrodes separately for each subject and condition.

Statistical analyses
To capture potential differences between different age groups in our sample, we used age as a factor with two levels.Because 13 years represents the median age of onset for SAD (Kessler et al., 2005), we used this age to split the sample, which resulted in subgroups of similar size (younger age group: 10-12 years old, n = 82, M = 11.6 years, SD = 0.89; older age group: 13-15 years old, n = 77, M = 14.4 years, SD = 0.83).A univariate analysis of variance (ANOVA) with Group (SAD, SP, HC) as the predictor assessed group differences in mean subjective arousal, which represents the average of the subjective state arousal levels rated before and after the EEG experiment.For behavioral measures, we calculated individual mean RTs for correct trials with RTs between 150 ms and 3 SDs above the mean (inclusion of 94 % of trials).A 3 (Group) x 2 (Age) x 3 (Condition) mixed-design ANOVA was calculated for behavioral measures (RTs and accuracy) in response to emotional faces.We used two additional 3 (Group) x 2 (Age) mixed-design ANOVAs to assess group differences in RTs and accuracy in response to objects.Regarding ERP data, we computed mixed-design ANOVAs to assess the effects of the within-factor Condition and the between-factors Group and Age on neural outcomes (P100, N170, EPN, and LPP) in emotional face trials.Regarding the LPP, we computed two separate models (early and late time window).For neural responses to non-social objects, we computed separate 3 (Group) x 2 (Age) mixed-design ANOVAs for each ERP (P100, N170, EPN, and LPPs).Because objects served mainly as control stimuli, we report effects for objects only when they involved the predictor Group.We ran all statistical analyses in RStudio (Posit team, 2023) with a significance level of α = .05;we applied Greenhouse-Geisser corrections if applicable and the corrected p-values and degrees of freedom are reported.Post-hoc Tukey tests for pairwise comparisons, which are robust to multiple testing, were calculated for significant main or interaction effects.To gain a comprehensive understanding of the results, we included additional trend level analyses of α < .10 for post-hoc analyses.Partial eta square (η p 2 ) represents the effect sizes of the ANOVAs.

Sample characteristics
Sample characteristics, including comorbid diagnoses, are displayed in Table 1.As predicted, youth with SAD had significantly higher mean scores in social anxiety symptoms, depressive symptoms, phobic fears, parent-reported general psychopathology, and in internalizing symptoms compared to both control groups.Also, as predicted, youth with SP significantly differed from HCs regarding phobic fears, parent-reported overall psychopathology, and internalizing symptoms.Groups did not differ in age, gender, or cognitive abilities.

P100
All mean ERP amplitudes to angry, happy, and neutral faces per Group and Age can be found in Table A1 of the appendix.The 2 (Group) x 2 (Age) x 3 (Condition) ANOVA revealed significant main effects of Age, F(1,153) = 28.43,p < .001,η p 2 = .157,and Condition F(1.94, 297.30) = 18.90 p < .001,η p 2 = .110.All other main or interaction effects did not reach significance, Fs < 1.94, ps > .14.The younger age group showed higher P100 amplitudes than the older one.Across groups, the presentation of angry and neutral faces led to increased P100 amplitudes compared to happy faces, t(153) = 6.571, p < .001,and t (153) = 4.071, p < .001.

N170
For the N170, there were significant main effects of Group, F( 2   (depicted in Fig. 4).No other main or interaction effects were significant, Fs < 2.51, ps > .085.Post-hoc comparisons revealed that only younger but not older participants with SAD showed lower EPN amplitudes to neutral, t(153) = 2.829, p = .015and happy faces, t(153) = 2.703, p = .021compared to younger HCs.In addition, younger participants with SAD showed a trend for lower EPN amplitudes in response to angry faces compared to younger HCs, t(153) = 2.251, p = .066.Other comparisons did not reach significance or trend level, ts( 153) < 1.709, ps > .20.

Early LPP
For the early time window of the LPP, the mixed ANOVA revealed a significant main effect of Age, F( 1 (153) < 2.043, ps > .11.

Control condition: response to non-social household objects
The mixed 2 (Group) x 2 (Age) ANOVAs for objects showed that behavioral measures (RTs and accuracy) were not influenced by Group or its interaction with Age, Fs < 1.97, ps > .14.Regarding neural responses (P100, N170, EPN, and LPPs), there were also no effects of Group or its interaction with Age, Fs < 1.88, ps > .15.

Exploratory analyses of age effects in the SAD group
To further investigate age differences in ERP results related to SAD, we conducted exploratory analyses regarding psychometric measures and ERP amplitudes.Younger and older participants with SAD did not significantly differ in raw scores on psychometric measures of psychopathology (ps >.37), with the exception of FSSC-R total scores related to phobic fears.Younger youth with SAD had significantly higher total scores (Mdn = 68) than older youth with SAD (Mdn = 55.5),W = 550, p = .020.Age effects related to psychometric measures were not evident in the SP (ps > .070)or HC group (ps > .18).To follow up on age differences in FSSC-R scores in youth with SAD, we explored associations between FSSC-R total scores and ERP components with a significant moderation effect of age (EPN) separately for the younger and older diagnostic groups.EPN values were not correlated with FSSC-R total scores in the younger, r(30) = .28,p = .15,or older age group, r τ (30) = .09,p = .48.There were also no significant correlations between FSSC-R total scores and EPN values in younger or older groups with SP or HCs (rs <.20, ps >.36).

Discussion
The aim of this study was to investigate the neural response to emotional faces and non-social household objects in children and adolescents aged 10 to 15 years with SAD.Our clinical controls were youth in the same age range with SP and HCs.(1) We expected youth with SAD to show early attentional processing biases (P100, N170, and EPN) and (2) higher LPP amplitudes in response to emotional faces compared to youth with SP and HCs.(3) In addition, we also examined whether age moderated group effects.(4) Finally, we hypothesized that group differences would only be present in response to emotional faces and not non-social household objects.Overall, our hypotheses were partially confirmed.While there were no group differences regarding the P100 and later stages of neural processing (late LPP), we found lower N170 amplitudes in children with SAD compared to HCs, irrespective of emotion and age.Furthermore, our results provide novel evidence of altered neural processing of facial stimuli that is dependent on age differences in SAD: Younger children with SAD showed lower EPN amplitudes in response to neutral and happy faces compared to younger HCs, followed by higher early LPP amplitudes (only trend level).Finally, no group differences were found in the neural response to non-social household objects, which emphasizes the specificity of the group differences observed for social-emotional stimuli.

Lower early neural processing in SAD only in response to emotional faces
Our study is the first one known to us that shows altered neural processing of facial stimuli in early ERP components in SAD during childhood and adolescence.More specifically, while we did not find group differences in the P100, youth with SAD showed lower N170 amplitudes when compared to HCs in response to all emotional faces.These findings contrast with the assumption of early hypervigilance toward threat in social anxiety, as suggested by dominant cognitive models (Clark & Wells, 1995;Rapee & Heimberg, 1997).They are also not in line with findings from adult EEG studies that show enhanced P100 amplitudes in socially anxious individuals (for a review see Harrewijn et al., 2017).In addition, our findings are only partially consistent with previous studies on face processing in children with clinical SAD, which also found no difference in the P100 (Keil et al., 2018;Schwab & Schienle, 2017, 2018).In line with our findings, lower N170 amplitudes have already been observed in previous dot-probe studies in a non-clinical sample of socially anxious children (Wauthia et al., 2023) and adults with SAD (Mueller et al., 2009).Furthermore, in a study conducted by Wieser and Moscovitch (2015) socially anxious adults showed blunted N170 amplitudes compared to non-socially anxious individuals when neutral faces were paired with verbal context information, which may indicate lower structural face encoding.This corroborates well with findings from a magnetoencephalography study that found an under-activation of the right fusiform gyrus, an area associated with the N170, in adults with SAD compared to HCs (Riwkes et al., 2015).The authors interpreted this as a differential processing of facial information.While individuals with social anxiety may neglect high spatial frequency information required for in-depth analyses of facial traits, they may rely more on global low spatial frequency information needed to quickly process more unrefined emotional cues (Vuilleumier et al., 2003).Thus, social anxiety may be associated with a fast decoding of emotional expressions which may be necessary to detect potentially threatening social cues.
A similar explanation stems from the proposition that the N170 is related to the activity of face-and eye-selective neurons (Itier et al., 2007).While the former mainly respond to the configuration of the face and yield large N170 amplitudes when upright faces are presented, the latter are particularly responsive to isolated eyes or to disrupted face configurations (e.g., inverted faces or faces with contrast reversal).Attenuated N170 amplitudes in youth with SAD in response to emotional faces may thus represent a lower response of face-sensitive neurons indicative of less configurational face processing.This interpretation would fit well with findings from an eye tracking study on childhood SAD.Youth with SAD showed restricted visual scanning of emotional faces, which may hinder a holistic representation of faces (Kleberg et al., 2021).
Our results point toward SAD-specific processes regarding the N170.Overall, youth with SP did not differ in their neural response to emotional faces when being compared to youth with SAD or HCs.Similar results have been obtained in the comparison between children with SAD and children with mixed anxieties (Keil et al., 2018).At the same time, other studies found higher LPP amplitudes in children (Leutgeb et al., 2010) and adults (Michalowski et al., 2009) with spider phobia compared to controls but importantly only in response to phobia-relevant stimuli.It should be noted that these studies did not include social stimuli and thus do not allow for conclusions regarding the transdiagnostic quality of emotional face processing.Together these findings underpin the need for research on the transdiagnostic quality of neural threat processing, especially with larger samples, to detect potentially small group differences.
Interestingly, in our study group differences in the neural processing of emotional faces between youth with SAD and controls were not reflected in behavioral measures.As previously reported (Keil et al., 2018;Kujawa et al., 2015;Wauthia et al., 2023), groups did not differ in RTs or A.-L. Rauschenbach et al. accuracy.In line with our hypothesis, the absence of group differences in the neural response to non-social household objects points toward a specificity of neural processing biases to social stimuli in SAD.

Neural face processing in SAD: age as a moderator
Our results showed interesting age-specific group differences in the neural response to emotional faces between diagnostic groups that point to developmental shifts in processing biases.Notably, group differences in youth with SAD were moderated by age and condition regarding the EPN.In detail, younger children with SAD showed lower EPN amplitudes compared to younger HCs in response to neutral and happy faces, though this comparison only reached trend level for angry faces.These findings contrast with the very few previous studies, one of which suggests that social anxiety in adults is associated with higher EPN amplitudes to pictures of human gaze (Schmitz et al., 2012), and another suggest no group differences in childhood SAD in response to emotional faces (Keil et al., 2018).Note, however, that the latter study with children had a slightly different age range and a smaller number of trials than ours, which potentially led to a lower signal to noise ratio and greater variance.The lower EPN amplitudes in younger children with SAD in our sample may indicate less selective attention to emotional cues (Junghöfer et al., 2017;Schupp et al., 2004) and could thus reflect a continued pattern of attenuated processing of social-emotional information already evident in the previous N170 component.Similarly, blunted pupillary reactivity has been found in response to emotional faces in 10-to 13-year old children with SAD (Keil et al., 2018).This was also interpreted as lower cognitive-affective processing and an indication of avoidance of anxiety-related stimuli.Overall, there is a lack of research on the EPN in emotional face processing, particularly in children, which makes it an interesting area for future investigation.
Why might it be that younger children with SAD in particular show a different pattern of attenuated neural face processing?Several explanations are plausible.In line with the argument for a developmental approach, the moderation model by Field and Lester (2010) suggests that particularly young children show attentional biases towards threat, which are further influenced by developmental processes and individual factors, such as anxiety.Exploratory analyses have revealed that younger children with SAD reported more general fears than older children with SAD.While a wider range of fears in younger children are part of typical development, children with SAD seem to also show a broader pattern of psychopathology compared to adolescents with SAD (Beesdo et al., 2009;Rao et al., 2007).However, in the current sample, higher levels of general fears were not significantly associated with EPN amplitudes.Consequently, age-related effects observed in the EPN component may indicate more general developmental processes in SAD, rather than reflecting the trajectories of general levels of phobic fears.A possible alternative explanation relates to the characteristics of the stimuli.Adolescence is associated with a greater focus on peer relationships and peer evaluation (Oerter & Montada, 2002), which could lead to a different emotional salience of peer versus adult faces based on age.Because previous research suggests that children show differential neural activation when processing peer compared to adult emotional faces (Marusak et al., 2013), it would be interesting to examine whether the same age effects on the EPN and LPP would also be observed in response to faces of peers.
Surprisingly, we did not find significant group differences in early or late LPP amplitudes but a trend toward higher early LPP amplitudes in younger children with SAD in response to happy and neutral faces.These findings contrast with previous studies that showed significantly higher LPPs in response to emotional faces in samples of children with SAD (Schwab & Schienle, 2017, 2018) and a subsample of youth with SAD (Kujawa et al., 2015), which was commonly interpreted as an index of sustained information processing.At the same time, our findings replicate those by Keil et al. (2018) who reported similar LPP amplitudes in children with SAD, mixed anxieties, and controls.While discrepancies in the stimulus material and experimental task may help to explain these inconsistencies, differences in the age of the sample may also play an important role.Group effects on the LPP were found particularly in young samples; the mean age in the studies by Schwab & Schienle, (2017;2018) was 9 years and in the study by Kujawa et al. (2015) the youngest participants were 7 years old (the mean age of the SAD subgroup was not reported).In contrast, the mean age in our younger age group was around 11 years.Extending this potential explanation, in our sample only younger children with SAD showed a trend for higher early LPP amplitudes.This again suggests potentially altered attentional processing, particularly in younger children, which is further amplified by clinical social anxiety.
Overall, age effects in youth with SAD and its effects on emotional face processing are not sufficiently examined yet.To our knowledge this is the first study to find age related changes in the neural processing of emotional faces in childhood SAD.Evidence from behavioral studies indicates that threat biases increase with age in anxious children and youth (Dudeney et al., 2015).At the same time, other studies have found that younger children with anxiety show a threat bias, while older children show attentional avoidance (Carmona et al., 2015;Reinholdt-Dunne et al., 2012).On a neural level, brain maturational processes and increases in cognitive control are likely to play an important role in age-related changes regarding the neural response to emotional faces (Luna et al., 2004).Taken together, the current findings underline the importance of a developmental approach.Due to our age range from 10 to 15 years, we may not have fully captured potential developmental shifts in neural processing biases or potential non-linear trajectories, so investigating age effects seems to be an important endeavor for future studies.

Limitations and future directions
The results of the current study should be interpreted in light of several limitations.Our relatively narrow age range did not include young children or older adolescents with SAD and thus may have inhibited us in capturing age effects in neural face processing from childhood to late adolescence.We focused specifically on age effects and did not examine potential effects of participant gender.In addition, we measured neural responses to static pictures of adult faces and findings may not generalize to real life social interactions.Youth with SAD reported higher state subjective arousal than control groups.Though statistical analyses showed no association between state subjective arousal and ERP components, we cannot rule out a possible effect on our results.Future studies would benefit from using different stimuli, such as comparing adult and child faces, using stimuli in motion, or examining social interactions (Lidle & Schmitz, 2022).Longitudinal studies with even broader age ranges would be helpful to further elucidate potential developmental trajectories.Another interesting endeavor would be to pair eye tracking with EEG to simultaneously capture the visual and neural processing of emotional faces.

Conclusion
Overall, this study showed that SAD in children and adolescents is associated with biased neural processing of social stimuli compared to HCs.In a rather large sample of youth aged 10 to 15, we found for the first time that SAD was associated with lower N170 amplitudes.Additionally, younger children with SAD showed a pattern of lower EPN amplitudes followed by higher early LPP amplitudes; the latter only reaching trend level.These processing biases seem to be specific to social stimuli and did not transfer to non-social household objects.The findings may inform our understanding of the development and maintenance of childhood SAD and emphasize the necessity of developmental approaches.Future research is required to further investigate age effects in neural face processing in SAD from childhood to late adolescence.

Fig. 2 .
Fig. 2. Scalp distribution of ERP components reflective of the response to faces averaged across all participants and conditions for the P100, N170, and LPP.For the EPN, the scalp distribution illustrates the difference between emotional (happy and angry) and neutral faces averaged across all participants.The circles represent electrodes included for each ERP.

Fig. 3 .
Fig. 3. Illustrating the grand mean of the N170 component in response to emotional faces averaged across P7 and P8 electrodes.SAD = Social Anxiety Disorder; SP = Specific Phobia; HC = Healthy Control.
Fig. 4. The interaction between Group, Age and Condition on grand average EPN amplitudes (averaged across POz, PO3, and PO4).SAD = Social Anxiety Disorder; SP = Specific Phobia; HC = Healthy Control.