High social anxiety is associated with attenuated cognitive-control modulation in a mixed threat–neutral context: evidence from ERPs and time-frequency analysis

Abstract

High social anxiety has been linked to impaired attentional control, yet its influence on stimulus-sensitive cognitive-control modulation in mixed threat-neutral contexts remains unclear. The present study examined cognitive-control-related behavioral and neural processing in individuals with high social anxiety (HSA) during an emotional Stroop task in which neutral and social threat words were randomly intermixed. Behaviorally, the HSA group responded more slowly than the low social anxiety (LSA) group, despite comparable accuracy across groups and conditions. In addition, responses to threat words were faster than those to neutral words across groups. At the neural level, neutral words elicited larger N2 and N450 amplitudes than threat words, and these differentiation effects were more pronounced in the LSA group than in the HSA group. Likewise, the LSA group showed clearer late-stage modulation in SP, whereas this effect was attenuated in the HSA group. In the time-frequency domain, theta power was stronger in the LSA group than in the HSA group, and neutral words elicited stronger alpha desynchronization than threat words in the HSA group. Taken together, these findings suggest that HSA is associated with reduced processing efficiency and attenuated stimulus-sensitive cognitive-control modulation in a mixed threat-neutral context.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04996-3.

Introduction

Cognitive control refers to the set of processes that enable the flexible allocation of cognitive resources and support goal-directed behavior [28, 29]. Accordingly, cognitive control has become a central topic in research on psychopathology and mental health [10, 12, 31]. Among its core components, inhibitory control and conflict monitoring are particularly important and may also have potential value for clinical assessment [40]. Social anxiety, also known as social phobia, is characterized by intense fear of negative evaluation in social situations, often accompanied by excessive concerns about embarrassment, humiliation, or being judged unfavorably by others, which in turn disrupt everyday social functioning [1]. Consistent with this view, previous studies have suggested that individuals with social anxiety (SA) show impairments in cognitive control [5, 35, 44].

Although numerous studies have investigated emotional processing in social anxiety using neuroimaging techniques such as fMRI and EEG, the temporal dynamics of cognitive control in socially anxious individuals remain insufficiently understood. To address this gap, the present study employed an emotional Stroop task and recorded event-related potentials (ERPs) to examine the time course of cognitive-control-related processing [47]. Owing to their high temporal resolution, ERPs are particularly well suited to capturing the dynamic stages of cognitive processing. Previous research has shown that cognitive control is closely associated with ERP components such as N2, N450, and SP (slow potential).

Numerous studies have shown that response inhibition and conflict processing are reflected in the N2 component over anterior-central scalp regions, and that N2 amplitude is closely associated with cognitive control [20]. Impaired cognitive control has been identified in a range of conditions, including nicotine dependence and social anxiety, making N2 a useful index for assessing such deficits [4, 44]. In a cognitive conflict experiment, Kanske and Kotz (2010) found that the N2 was sensitive to the level of emotional conflict, suggesting that this component can be modulated by emotional information. Using an affective Stroop task, Yang et al., [49] found that neutral stimuli elicited larger N2 amplitudes than negative stimuli, suggesting that negative emotion was prioritized and consumed more cognitive resources than neutral emotion [49]. Their results also revealed stronger top-down cognitive-control connectivity from frontal to occipital regions under negative emotional conditions, further indicating that emotional information can influence cognition by consuming control resources and increasing the need for cognitive control during task performance. Furthermore, N450 is widely regarded as an important electrophysiological marker of interference processing in cognitive control. Previous studies have shown that N450 amplitude increases as stimulus interference becomes stronger, highlighting its role in conflict monitoring and interference suppression [9, 39]. In the Stroop task, Nixon et al., [36] found that sadness enhanced the interference effect while reducing N450 amplitude, suggesting that negative emotion impaired conflict monitoring and hindered the allocation of cognitive resources to the target task [36]. In addition, Cao et al., [5] reported that individuals with low social anxiety (LSA) could effectively monitor and resolve conflict during task performance, whereas those with high social anxiety (HSA) showed reduced ability to do so. Consistent with this view, previous work has also shown that the amplitude and topographic distribution of electrophysiological markers can distinguish between healthy and clinical populations, including individuals with attention-deficit/hyperactivity disorder [30]. In previous studies, SP amplitudes have been found to be more positive under high-conflict than under low-conflict conditions, suggesting that SP is involved in conflict resolution and later-stage cognitive-control adjustment [28, 41]. In an emotional Stroop task, Cao et al., [5] reported that conflict-related SP amplitudes were more positive under incongruent than congruent conditions in HSA participants, but not in LSA participants, suggesting altered control-related processing in socially anxious individuals [5]. By contrast, Qi [41], found that conflict-related SP modulation was evident in the low trait anxiety (LTA) group but absent in the high trait anxiety (HTA) group. Together, the existing evidence remains somewhat mixed, and further research is needed to clarify how anxiety modulates SP during conflict processing.

Theta-band oscillations, which are thought to originate primarily from the medial prefrontal cortex (mPFC), play an important role in top-down cognitive control and have been found to be negatively associated with anxiety levels [18, 33]. By contrast, alpha-band activity is commonly linked to the suppression of task-irrelevant information and is observed in a wide range of cognitive tasks requiring the maintenance of attentional resources [16, 21]. Previous studies have also shown that attention to a stimulus is often accompanied by reduced alpha power, reflecting active information processing such as attention and memory [48].

Social threat stimuli, as disorder-relevant materials, may better capture the concerns of individuals with social anxiety and therefore provide a more sensitive means of examining cognitive-control-related processing in this population [34, 47]. For this reason, the present emotional Stroop task included both social threat and neutral words. In a mixed threat–neutral context, participants cannot predict whether the upcoming stimulus will contain socially threatening content. This contextual uncertainty may promote sustained monitoring of threat-related information while participants must still maintain the task goal of identifying word color. Such a context may be disproportionately demanding for individuals with high social anxiety (HSA), because social threat cues are closely related to fear of negative evaluation and may therefore compete more strongly with goal-directed attentional control. As a result, even neutral words may not be processed in a functionally threat-free manner when they are embedded in a context where social threat is continuously possible.

In the present study, we use the term “stimulus-sensitive cognitive-control modulation” to refer to the degree to which control-related neural responses were modulated by stimulus type within the same mixed threat-neutral context. Specifically, stronger differentiation between social threat and neutral words in ERP amplitudes and oscillatory power was interpreted as greater stimulus-sensitive modulation of cognitive control, whereas attenuated differentiation was interpreted as reduced control-related modulation. Importantly, the present study did not directly assess behavioral task-switching flexibility or trial-by-trial adaptive flexibility. Therefore, our use of “modulation” refers specifically to stimulus-sensitive changes in control-related neural activity. Because the present study used an extreme-group design comparing HSA and low social anxiety (LSA) participants, our primary expectation concerned categorical group differences rather than a specific linear or non-linear association across the full social-anxiety continuum. We expected that, relative to LSA participants, HSA participants would show reduced processing efficiency and attenuated stimulus-sensitive modulation of cognitive-control-related neural activity in the mixed threat–neutral context. Future studies using full-range samples are needed to determine whether the relationship between social anxiety severity and cognitive-control deficits is linear, threshold-like, or otherwise non-linear.

While the same task and sample were used in our previous study [6], which focused on emotion-processing effects, the present study addresses a distinct question by examining stimulus-sensitive cognitive-control-related modulation in a mixed threat-neutral context using additional analytical approaches, including previous-trial analysis, SP modulation, time-frequency analysis, and differentiation-score analyses.

Method

Participant

Participants were recruited through WeChat Moments and asked to fill in the Chinese version of the Liebowitz Social Anxiety Scale (LSAS) online [19, 26]. Individuals with scores greater than 60 are defined as HSA [27], and those with scores less than 35 are defined as LSA [38]. 42 LSA and 42 HSA participants were invited to the lab to take part in the experiment. Each participant had a normal or corrected-to-normal vision. Five participants were excluded from the EEG analyses because more than 25% of their trials were rejected due to EEG artifacts after preprocessing. This participant-level threshold was used to ensure sufficient artifact-free trials for each condition and stable ERP averaging. In addition, two participants were excluded because their overall task accuracy was below 75% (the mean accuracy was 0.71 for threat conditions and 0.73 for neutral conditions). Consequently, a total of 77 subjects (mean age: 20.7 ± 1.9; 60 females, 17 males) were included, consisting of 38 HSA and 39 LSA. For the HSA and LSA subjects, the LSAS were 77.13 ± 15.94 and 20.31 ± 10.65, respectively; t (75) = 18.439, p < .001. After receiving ethical approval from the Ethics Committee of Daqing Campus of Harbin Medical University (No.: HMUDQ20240420103) following the Declaration of Helsinki, all participants signed informed consent forms.

Material

The stimuli used in this study were identical to those used in the previous study [6, 44]. 30 social threat words and 30 neutral words were selected from the social threat word system compiled by the authors. Each word was rated by an independent sample of 36 Chinese speakers on four dimensions: valence, arousal, familiarity, and social relevance. For more detailed information on the stimuli and rating process, please refer to Cao et al., [6].

During the emotional Stroop task, the experimental design and procedure were identical to those described in our previous study [6]. Briefly, participants sat 70 cm from a computer monitor, and stimuli were presented with a visual angle of 5.2◦×2.3◦. As shown in Fig. 1, each trial began with a fixation cross (500–700 ms), followed by an emotional word (800 ms), and a blank screen (700 ms). A total of 240 trials were presented in a single mixed sequence, including 120 neutral-word trials and 120 social threat-word trials. The two types of words were intermixed throughout the task and presented in a randomized order for each participant, rather than in separate emotion-specific blocks. Thus, participants could not predict whether the upcoming trial would contain a neutral or social threat word, which operationally defined the mixed threat-neutral context in the present study. Participants identified the color of the words using keyboard response keys. For further information, please see the work by Cao et al., [6].

Fig. 1.

Sample stimuli and procedures for the emotional Stroop task

The present experiment was a mixed design with social anxiety (high and low) as a between-subject factor and emotion (neutral and threat) as a within-subject factor.

EEG recording and processing

EEG data were recorded using a Neuroscan ESI 64-channel system with an Ag/AgCl electrode cap following the extended 10–20 system. The sampling rate was 500 Hz, with a left mastoid reference electrode used online and re-referenced offline to the average of left and right mastoids. Electrooculogram (EOG) signals were recorded using electrodes placed vertically and horizontally around the eyes. Impedances were kept below 5 kΩ throughout the experiment. The EEG signals were bandpass filtered between 0.05 Hz and 30 Hz.

EEG data were segmented into 1000 ms epochs, starting 200 ms prior to stimulus onset, and baseline corrected using the 200 ms pre-stimulus period. Trials with voltages exceeding ± 75µV were excluded. EOG artifacts were corrected using a linear regression [43]. The average ERPs were computed from correct-response trials. Data preprocessing was performed with Neuroscan Scan 4.5 software, and statistical analysis was conducted using SPSS 29.0.

For the TFR analyses, EEG data were segmented for 2000 ms starting 500 ms before stimuli onset. Waveforms were corrected relative to the baseline period of 500 ms before the stimulus. Before averaging, trials with EEG voltages exceeding ± 75µV were removed. A linear regression estimate was used to correct EOG blink artifacts.

Statistics and analyses

Behavioral analyses

Accuracy was analyzed using a generalized linear mixed model (GLMM) with Group (HSA, LSA), Emotion (neutral, threat), and their interaction entered as fixed effects, and Subject as a random intercept. Reaction time (RT) analyses were conducted on correct trials only. Trials with RTs shorter than 200 ms and trials exceeding ± 3 SD from each participant’s mean within each emotion condition were excluded. Because RT data were positively skewed, log-transformed RTs were analyzed using linear mixed-effects models (LMMs), with Group, Emotion, and their interaction entered as fixed effects and Subject as a random intercept. To further examine control-related performance for neutral-word processing in a mixed threat–neutral context, a focused analysis was conducted on current neutral trials, in which Previous-trial Emotion (neutral, threat), Group, and their interaction were entered as fixed effects. In addition, RT variability was indexed by the coefficient of variation (CV = SD/mean RT) for each participant under each emotion condition and was analyzed using a linear mixed-effects model with Group, Emotion, and their interaction as fixed effects and Subject as a random intercept.

ERP analyses

Based on inspection of the grand average ERP waveforms (Fig. 2) and previous studies [5, 11, 25], the mean amplitudes were computed at time windows of 250–320 ms (for N2), 380–480 ms (for N450), and 480–800 ms (for SP). The mean amplitude during each time window was obtained from five electrode sites: Fz, FCz, Cz, CPz, and Pz. A three-way repeated-measures ANOVA was conducted for the measurements of N2, N450, and SP with the following factors and levels: 2 (Group: HSA, LSA) × 2 (Emotional type: neutral, threat) × 5 (Site: Fz, FCz, Cz, CPz, and Pz).

Fig. 2.

A Grand average ERP waveforms under threat and neutral conditions in HSA and LSA groups; B Topographic map of difference amplitudes (subtracting ERPs to neutral words from ERPs to threat words) across three intervals (250–320 ms, 380–480 ms, and 480–800 ms post stimuli onset)

The reported p-levels of all repeated-measured ANOVAs were adjusted for violations of the sphericity assumption using Greenhouse-Geisser correction. The alpha level of significance was set as 0.05. If the interaction effect was significant (p < .05), simple effects were explored, and all pairwise comparisons were adjusted by a Bonferroni procedure.

TFR analyses

On the basis of previous studies [7, 24], for the sake of comparison, we focus on the same frequency bands across studies: theta band (4–6 Hz) and alpha band (10–12 Hz). Fast Fourier Transform (FFT) implemented in Fieldtrip was applied to compute time-frequency representations (TFRs) of power [37]. To minimize spectrum leakage, a Hanning taper was applied prior to FFT. A 500 ms sliding window with 90% overlap was used to enhance temporal resolution, and power values were normalized relative to the pre-stimulus baseline (-400 ms to -100 ms). Data normalization was performed using decibel (dB) conversion [dB power = 10 log (power/baseline)].

Following visual inspection of the spectrogram and in line with previous research [8, 16], we calculated the mean power oscillations in the theta band (350–550 ms) for the frontal and fronto-central electrodes (F1, Fz, F2, FC1, FCz, and FC2), and in the alpha band (300–450 ms) for the parietal and parieto-occipital electrodes (P1, Pz, P2, PO3, POz, and PO4).

A three-way repeated-measures ANOVA was conducted for the measures of the theta band with the following factors and levels: 2 (Social anxiety: high, low) × 2 (Emotional type: neutral, threat) × 6 (electrode: F1, Fz, F2, FC1, FCz, FC2). A three-way repeated-measures ANOVA was conducted for the measures of the alpha band with the following factors and levels: 2 (Social anxiety: high, low) × 2 (Emotional type: neutral, threat) × 6 (electrode: P1, Pz, P2, PO3, POz, PO4).

Differentiation-score analyses

To directly quantify stimulus-sensitive neural modulation, differentiation scores were calculated for each participant by subtracting the mean response to neutral words from the mean response to threat words. Thus, threat-minus-neutral difference scores were computed for N2, N450, SP, theta power, and alpha power. For N450, SP, theta, and alpha power, scores were averaged across the predefined electrode clusters before difference scores were calculated. For N2, because the Emotion × Site × Group interaction was significant, site-specific difference scores were additionally examined across Fz, FCz, Cz, CPz, and Pz. Group differences in differentiation scores were tested using independent-samples t tests, and within-group differentiation effects were examined using one-sample t tests against zero.

Results

Behavioral data

Accuracy was analyzed using a generalized linear mixed model. The results showed no significant main effect of group, β = -0.003, SE = 0.161, z = -0.02, p = .986, no significant main effect of emotion, β = 0.131, SE = 0.101, z = 1.29, p = .196, and no significant Group × Emotion interaction, β = -0.101, SE = 0.142, z = -0.71, p = .476, indicating comparable task accuracy across groups and emotion conditions. For log-transformed RTs, the linear mixed-effects model revealed a significant main effect of group, F_{(1, 75)} = 5.43, p = .022, with slower responses in the HSA group than in the LSA group. A significant main effect of emotion was also observed, F_{(1, 16488)} = 4.02, p = .045, indicating overall faster responses to threat than to neutral words. The Group × Emotion interaction was not significant, F_{(1, 16488)} = 0.27, p = .606. Estimated marginal means showed numerically slower responses in the HSA group than in the LSA group under both neutral and threat conditions. Tukey-adjusted pairwise comparisons among the four Group × Emotion cells are reported in Supplementary Table S2. Only the contrast between HSA-neutral trials and LSA-threat trials reached significance, indicating that the slowest responses were observed for neutral trials in the HSA group relative to threat trials in the LSA group. All other pairwise comparisons were not significant.

To further examine control-related performance, current neutral trials were analyzed with previous-trial emotion included as a factor. This analysis revealed a significant main effect of group, F_{(1, 75.2)} = 5.42, p = .023, indicating that HSA participants responded more slowly than LSA participants. The main effect of previous-trial emotion did not reach statistical significance, F_{(1, 8200.5)} = 2.73, p = .099, and the Group × Previous-trial Emotion interaction was also not significant, F_{(1, 8200.5)} = 2.61, p = .107. Thus, this focused analysis did not provide strong evidence for a reliable immediate carryover effect from the preceding trial, nor for a group-specific previous-trial effect in HSA participants. In addition, analysis of RT variability showed no significant main effect of group, F_{(1, 75)} = 2.07, p = .154, no significant main effect of emotion, F_{(1, 75)} = 2.03, p = .158, and no significant Group × Emotion interaction, F_{(1, 75)} = 0.28, p = .595.

ERP data

N2 (250–320 ms)

There was no significant main effect of Group, F < 1, indicating that overall N2 amplitudes did not differ significantly between the HSA group (M = 3.7 µV) and the LSA group (M = 3.1 µV). The three-way ANOVA showed a significant main effect of Emotion type, F_(1,75) = 101.275, p < .001, = 0.575, revealing that neutral words elicited more negative going N2 (2.7 uV) than did threat words (4.2 uV). The main effect of the Site was also significant, F_(4,300) = 10.323, p < .001, = 0.121, showing the frontal distribution with the most negative N2 at Fz site (2.7 uV). Both the interactions of Emotion type × Site [F (4,300) = 28.186, p < .001, = 0.273] and Emotion type × Site × Group were significant [F_(4,300) = 5.517, p < .001, = 0.069]. Follow-up analyses were conducted to further examine the significant Emotion × Site × Group interaction. Simple-effect analyses showed that the Emotion effect was significant at all five midline sites in both groups, with neutral words eliciting more negative-going N2 amplitudes than threat words. Full statistics are provided in Supplementary Table S3, Part A.

To further identify the site at which the group difference in emotion-related N2 differentiation was most pronounced, threat-minus-neutral differentiation scores were compared between groups at each midline site. As shown in Supplementary Table S3, Part B, the group difference was most evident at Fz, where the LSA group showed a larger differentiation score (N2_threat minus N2_neutral: 2.2 uV) than the HSA group (N2_threat minus N2_neutral: 1.4 uV), t₍₇₅₎ = -2.05, p = .044, Cohen’s d = -0.47. Although this site-wise comparison should be interpreted cautiously because it did not survive correction for multiple comparisons, it suggests that anterior N2 differentiation was more pronounced in the LSA group than in the HSA group (see Fig. 2).

N450 (380–480 ms)

Similar to the ANOVA analysis of N2 amplitudes, for N450 amplitudes, the group difference was not significant, F < 1 (5.8 uV and 5.5 uV for HSA and LSA groups, respectively). There was a significant main effect of Emotion type, F_(1,75) = 25.164, p < .001, = 0.251, revealing that neutral words elicited more negative going N450 (6.2 uV) than did threat words (5.2 uV). The Emotion × Group interaction was significant, F_(1,75) = 6.601, p = .012, = 0.081. Simple-effect analyses showed that the emotion effect was significant in the LSA group, threat minus neutral = 1.4 µV, p < .001, whereas the corresponding effect in the HSA group did not reach significance, threat minus neutral = 0.4 µV, p = .090. This pattern suggests attenuated N450 modulation in HSA, although the within-group effect did not reach statistical significance. The main effect of Site was also significant, F_(4,300) = 35.658, p < .001, = 0.322, showing the frontal distribution with the most negative N450 at Fz site (4.2 uV). No other effect reached a significant level (ps > 0.1); see Fig. 2.

SP (480–800 ms)

The three-way ANOVA showed that the threat words elicited more positive SP (4.5 uV) than did neutral words (3.9 uV), F_(1,75) = 13.044, p < .001, = 0.148. Although the group difference was not significant (4.6 uV and 3.7 uV for HSA and LSA groups, respectively; F_(1,75) = 1.693, p = .197, = 0.022), the Emotion × Group interaction was significant, F_(1,75) = 5.931, p = .017, = 0.073. Simple-effect analyses showed a significant SP modulation in the LSA group, threat minus neutral = 0.9 µV, p < .001, whereas the corresponding effect in the HSA group was not significant, threat minus neutral = 0.18 µV, p = .411. This result suggests weakened late-stage control-related modulation in HSA and should therefore be interpreted with appropriate caution. The main effect of Site was also significant, F_(4,300) = 72.00, p < .001, = 0.699, showing the central-parietal distribution with the largest SP at Cz site (5.3 uV). No other effect reached a significant level (ps > 0.1); see Fig. 2.

Time-frequency analysis

Theta band (350–550 ms)

A three-way ANOVA revealed a significant main effect of Social anxiety [F_(1,75) = 7.967, p = .006, = 0.100], showing the theta power was stronger in the LSA group (M = 2.53, SE = 0.15) than in the HSA group (M = 1.92, SE = 0.15); see Fig. 3.

Fig. 3.

A The theta oscillatory activity of neutral and threat conditions in HSA and LSA groups, respectively; B The alpha oscillatory activity of neutral and threat conditions in HSA and LSA groups, respectively

Alpha band (300–450 ms)

A three-way ANOVA revealed a significant Emotion × Group interaction, [F_(1,75) = 4.188, p = .044, = 0.055]. Simple-effect analysis showed that, in the HSA group, neutral words elicited more negative alpha power values than threat words, indicating stronger alpha desynchronization, p = .001; this effect was not significant in the LSA group, p = .603; see Fig. 3.

Additional differentiation-score analyses

Additional differentiation-score analyses were conducted to directly quantify stimulus-sensitive neural modulation. For N450, the LSA group showed a significantly larger threat-minus-neutral differentiation score than the HSA group, t₍₇₅₎ = -2.57, p = .012, Cohen’s d = -0.59. Similarly, the SP differentiation score was larger in the LSA group than in the HSA group, t₍₇₅₎ = -2.44, p = .017, Cohen’s d = -0.56. For theta power, the group difference was not significant, whereas for alpha power, the HSA group showed a larger threat-minus-neutral differentiation score, reflecting stronger alpha desynchronization for neutral words. Full differentiation-score results are provided in Supplementary Table S1.

Discussion

The present study examined how HSA shapes cognitive-control-related processing during an emotional Stroop task in which neutral and social threat words were randomly intermixed. At the behavioral level, participants with HSA responded more slowly than those with LSA, despite comparable accuracy across groups and conditions. At the neural level, the LSA group showed clearer differentiation between neutral and threat words across N2, N450, and SP, whereas this modulation was attenuated in the HSA group. A similar group contrast emerged in the time-frequency domain. Theta activity was stronger in the LSA group, whereas neutral words elicited stronger alpha desynchronization than threat words in the HSA group. Viewed together, these findings suggest that HSA is associated less with a simple amplification of threat responses than with reduced processing efficiency and attenuated stimulus-sensitive modulation of control-related neural activity in a mixed threat-neutral context.

The mixed threat–neutral context is important for interpreting the present findings. Because threat and neutral words were randomly intermixed, participants had to maintain the color-naming goal while continuously filtering potentially salient social-threat information. This context may be especially demanding for HSA individuals because social-threat cues are disorder-relevant and may increase sustained vigilance. Under such conditions, neutral words may not function as purely neutral stimuli; rather, they are processed within a task environment in which threat is repeatedly possible. This may help explain why HSA was associated with slower responding and attenuated stimulus-sensitive neural modulation, especially when neutral and threat information had to be processed within the same task context.

The behavioral pattern points primarily to reduced processing efficiency rather than to a speed-accuracy tradeoff. Accuracy remained stable across groups and emotion conditions, yet the HSA group showed slower overall responses, and threat words were named faster than neutral words across groups. This pattern supports attentional control theory, which argues that anxiety tends to impair processing efficiency even when overt performance can still be maintained through compensatory effort [3, 13]. In the present task, participants had to identify word color while suppressing task-irrelevant semantic content. Under such conditions, socially anxious individuals may need to invest more effort to keep performance on track, which would be expected to slow responding without necessarily reducing accuracy [2, 14, 50]. Notably, the focused analysis of current neutral trials showed the same group difference, whereas previous-trial emotion showed only a marginal effect and did not significantly interact with group. The previous-trial analysis did not provide strong evidence for a reliable group-specific sequential carryover effect. Therefore, the slower responses observed in the HSA group are more cautiously interpreted as reflecting reduced processing efficiency in the mixed threat–neutral context rather than a reliable HSA-specific carryover from the immediately preceding trial. Because the same task and dataset were previously used to address emotion-processing effects, the behavioral findings here are best read as supportive evidence for a cognitive-control account rather than as a separate demonstration of emotional bias. In this respect, the present study extends the earlier report by showing that slower performance in HSA is accompanied by altered control-related neural dynamics [5, 6].

At the N2 stage, neutral words elicited larger negative-going amplitudes than threat words across groups. N2 is commonly linked to executive processes in Stroop-like tasks, including conflict monitoring, inhibition recruitment, and the allocation of attention to task-relevant stimulus features [15, 17, 20]. Seen from that perspective, the larger N2 for neutral words may indicate that these trials required stronger enhancement of goal-relevant information, namely color, in order to overcome interference from word meaning. Threat words, by contrast, may have benefited from their motivational salience and therefore required less additional control recruitment during color naming, a pattern that aligns with reports of larger N2 responses to neutral than threat stimuli in affective conflict paradigms [49]. The frontal distribution observed here is also in line with earlier work linking N2 to frontal control-related mechanisms [5, 15].

More importantly, the contrast between neutral and threat words was sharper in the LSA group than in the HSA group. Rather than showing stronger threat-specific enhancement, the HSA group showed attenuated early-stage differentiation between stimulus types. This pattern can be understood in relation to attentional control theory, which proposes that anxiety reduces the efficiency of goal-directed attentional control by increasing the influence of stimulus-driven processing, especially when threat-related information competes with task goals [12, 13]. In the present mixed threat-neutral context, HSA participants may have maintained a more globally heightened state of vigilance for socially evaluative information. Such sustained monitoring may have reduced the selectivity with which early control resources were allocated to neutral and threat trials. Thus, the attenuated N2 differentiation in HSA may reflect less efficient or less selective early control recruitment in a mixed threat–neutral context, rather than a simple enhancement of initial threat sensitivity.

The N450 results extended this pattern into a later stage of conflict monitoring. Neutral words elicited larger negative-going N450 amplitudes than threat words, but this emotion-related modulation was significant in the LSA group and did not reach significance in the HSA group. In Stroop-type tasks, N450 is widely regarded as an index of conflict monitoring and interference processing. Emotional stimuli, because of their salience, may receive prioritized processing and thereby reduce conflict-related demands, leading to smaller N450 amplitudes than those elicited by neutral stimuli [22, 23, 42, 45]. The present findings fit this logic in the LSA group, where threat words may have been processed more efficiently and therefore generated less conflict-related monitoring than neutral words. By contrast, the HSA group showed attenuated N450 differentiation between neutral and threat words. Together with the significant Emotion × Group interaction and the differentiation-score analysis, this pattern suggests weaker stimulus-sensitive modulation of conflict-monitoring processes in HSA relative to LSA. According to attentional control theory and broader cognitive-control accounts of anxiety, anxiety can reduce the efficiency of goal-directed control when task-relevant processing must be maintained in the presence of emotionally salient distractors [2, 12, 13]. In the present mixed threat-neutral context, HSA participants may have engaged conflict-monitoring resources in a less selective manner across neutral and threat trials, possibly because social-threat information remained continuously relevant. Thus, the N450 findings suggest attenuated stimulus-sensitive conflict-monitoring modulation in HSA.

A comparable attenuation of stimulus-sensitive differentiation was also observed in the SP results, but this effect appeared at the level of later-stage control adjustment. Threat words elicited more positive SP amplitudes than neutral words in the LSA group, whereas this emotion-related modulation did not reach significance in the HSA group. SP has often been interpreted as reflecting later control operations such as response selection, conflict resolution, and the sustained implementation of control after conflict has been detected [46]. It has also been linked to supramodal attentional control systems that support goal-relevant processing while suppressing distraction from irrelevant input [11]. Against this background, the present pattern suggests that LSA participants adjusted late-stage control processes in a stimulus-sensitive manner, whereas HSA participants showed attenuated SP differentiation between neutral and threat words. This pattern should be interpreted as weaker stimulus-sensitive late-stage control adjustment in HSA. Together with the significant Emotion × Group interaction and the differentiation-score analysis, the SP findings suggest attenuated stimulus-sensitive late-stage control modulation in HSA.

Taken together, the ERP findings suggest that HSA was associated with attenuated stimulus-sensitive modulation across multiple stages of cognitive control, including early attentional control, conflict monitoring, and late-stage control adjustment. Importantly, these results suggest weaker, rather than absent, control-related neural modulation in HSA participants. Specifically, neural responses were less differentiated between neutral and social threat words in a mixed threat-neutral context. This pattern is consistent with cognitive-control accounts of anxiety and with hypervigilance theories, according to which sustained monitoring for threat may reduce the efficiency and selectivity of goal-directed control.

The oscillatory results converged with the ERP findings while adding another perspective on control-related processing in the mixed threat–neutral context. Theta power was stronger in the LSA group than in the HSA group, suggesting more efficient engagement of control-monitoring mechanisms in low-anxiety individuals. Frontal theta activity has repeatedly been linked to cognitive control, conflict monitoring, and the signaling of the need for top-down regulation during demanding tasks [8]. In conflict and Stroop-like tasks, stronger frontal theta is often interpreted as reflecting more effective recruitment of control resources when task-relevant goals must be maintained in the presence of interference. From this perspective, the reduced theta activity in the HSA group may indicate less efficient engagement of top-down control processes when neutral and social threat information had to be processed within the same task context. This interpretation is also consistent with previous reports of diminished control-related neural activity in anxiety [5, 24, 32].

The alpha-band finding should be interpreted in terms of alpha desynchronization rather than increased alpha power. In the HSA group, neutral words elicited more negative alpha power values than threat words, indicating stronger alpha desynchronization. Alpha desynchronization is often associated with active stimulus processing, attentional engagement, and the release of inhibition over task-relevant cortical regions. Thus, stronger alpha desynchronization for neutral words in HSA may suggest that neutral trials required greater task engagement or inhibitory-control demand when they appeared in a context where social threat was repeatedly possible. In this sense, neutral words may not have functioned as purely neutral stimuli for HSA participants; rather, they may have become more effortful to process because the task context maintained a sustained expectation of socially threatening information. This pattern fits the behavioral slowing observed in the HSA group and complements the ERP evidence of attenuated stimulus-sensitive neural differentiation.

Taken together, the behavioral, ERP, and TFR findings support a common conclusion. HSA was associated with less efficient processing and attenuated stimulus-sensitive modulation of control-related neural activity in a mixed threat-neutral context. The critical pattern was not simply stronger responding to threat. Instead, HSA participants appeared to show weaker differentiation between neutral and threat stimuli across multiple control-related measures, particularly when neutral information had to be processed in an environment that also contained socially threatening cues.

Clinical implications

Although the present study was not designed to test intervention effects, the findings may have preliminary clinical implications for social anxiety. The attenuated stimulus-sensitive modulation observed in HSA suggests that future intervention studies may consider whether goal maintenance and interference control in emotionally mixed contexts are useful treatment-related targets. However, these implications should be interpreted cautiously, as the present study was cross-sectional and did not directly test treatment mechanisms.

Limitations

Before concluding, several limitations should be acknowledged. First, the present study used social threat and neutral words as stimuli. Although these materials were appropriate for examining cognitive-control-related processing in an emotional Stroop task, they may not have evoked emotional responses as strongly as more ecologically valid social threat stimuli, such as facial expressions or social-evaluative scenes. Future studies may therefore benefit from using more naturalistic materials to further examine cognitive control under socially threatening conditions. Second, the stimulus presentation duration was 800 ms, which may have allowed relatively extended processing of word meaning and thus may have reduced sensitivity to more rapid emotion-related modulation of cognitive control. A shorter presentation duration might elicit more immediate emotional responses and provide additional insight into the rapid temporal dynamics of emotional interference and control adjustment. Future studies could systematically manipulate stimulus duration to determine whether the observed ERP and oscillatory patterns are more pronounced during earlier stages of emotional processing. Third, individual differences in coping strategies, emotion regulation, and avoidance tendencies were not assessed in the present study. Such factors may influence how individuals with HSA allocate cognitive-control resources under emotional interference. For example, participants who habitually use avoidance or suppression strategies may show different behavioral slowing, ERP differentiation, or oscillatory responses compared with those who use more adaptive regulation strategies. Future studies should include measures of coping style and emotion-regulation strategies to examine whether these individual differences moderate cognitive-control performance in mixed threat–neutral context. Fourth, threat and neutral words were randomly intermixed in the present design. Although this arrangement allowed us to examine performance in a mixed threat-neutral context, it also means that neutral-word processing could not be fully separated from the influence of threat-related context. Accordingly, the present findings should not be interpreted as reflecting neutral processing in a threat-free condition. Future studies should compare mixed and pure contexts more directly, such as pure-neutral, pure-threat, and mixed threat–neutral blocks, in order to better isolate the contribution of contextual threat, stimulus type, and emotional interference to cognitive-control performance. Finally, the present study used an extreme-group design and therefore could not determine whether cognitive-control alterations increase linearly with social anxiety severity or emerge in a threshold-like or non-linear manner. In addition, although the sample size was comparable to many ERP studies, no a priori power analysis was conducted specifically for the present differentiation-score analyses. Therefore, non-significant within-group effects, especially in the HSA group, should be interpreted cautiously. Future studies should recruit participants across the full range of social anxiety severity and use larger samples to test continuous linear, quadratic, or threshold models and to confirm the robustness of these effects.

Conclusion

The present findings suggest that HSA is associated with slower behavioral responding and attenuated cognitive-control-related neural differentiation during an emotional Stroop task with randomly intermixed neutral and social threat words. Across ERP and time-frequency measures, HSA was associated with weaker stimulus-sensitive modulation of control-related neural activity, particularly during conflict monitoring and late-stage control adjustment. These findings should be interpreted as evidence for reduced processing efficiency and attenuated neural modulation in a mixed threat–neutral context, rather than as a direct behavioral demonstration of impaired cognitive flexibility. Together, the results highlight that, in social anxiety, difficulty may lie not only in processing threat, but also in efficiently regulating attention and interference control in emotionally mixed contexts.

Authors’ contributions

H.W., X.Z. and Q.Y. collected the data and wrote the first draft of the manuscript. Y.Z. managed the literature searches and summaries of previous related work. M.G. designed the study and performed the statistical analysis. J.C. reviewed and revised the manuscript. All authors reviewed the manuscript.

Funding

The present study is supported by the Fundamental Research Funds for the Provincial Universities (JFJC202302), China, the Natural Science Foundation of Heilongjiang Province (LH2024H031), China, and the Humanities and Social Sciences Project, Ministry of Education (24YJAZH005), China.

Data availability

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The studies involving human participants were reviewed and approved by the Ethics Committee of Daqing Campus of Harbin Medical University (No.: HMUDQ20240420103) following the Declaration of Helsinki; all participants signed informed consent forms in this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.