Neural responses to error in youth: the impact of social context, anxiety, and worry

Abstract

Atypical error responses characterize pediatric anxiety disorders. Error responses are considerably influenced by situational factors such as social context, which typically elicits enhanced error responses. This study investigates the impact of social context on the neural correlates of error processing and how individual differences in anxiety symptoms and naturalistically-sampled daily worry influences these associations. Sixty-two youth (32 with an anxiety disorder, 30 healthy controls; M_age =13.9 +/- 2.7, 61% female) underwent functional magnetic resonance imaging while completing a flanker task in both peer and alone context. Anxiety symptoms were measured using Screen for Child Anxiety Related Disorders, and daily worry was measured using ecological momentary assessment. Diagnosis inclusions were only used to ensure enriched sample for symptoms dimensions of anxiety but not used as a grouping criteria. The presence of a simulated peer was associated with decreased activity in the precuneus, mid-orbital gyrus, and parahippocampal gyrus/amygdala during error processing. In youth with higher anxiety symptom severity, the presence of a peer was associated with decreased activity in the superior/middle temporal gyrus and middle frontal gyrus. In contrast, for youth rating higher levels of daily worries, the presence of a peer was associated with increased activity in the inferior frontal gyrus and middle cingulate cortex. Findings highlight the importance of social context in error processing. Results further suggest that anxiety and worry differentially modulate neural responses in the presence of a peer.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00787-025-02957-6.

Introduction

Error processing, how individuals evaluate their mistakes, underlies key facets of cognitive control [1, 2]. Atypical error processing relates to pediatric anxiety [1, 2]. Electrophysiological research finds aberrant error-linked, event-related potentials (ERPs), with recent work highlighting the importance of context, specifically peer observation [3, 4]. In contrast, the contribution of neural systems engaged by error processing in the context of social scrutiny remains underexplored with functional magnetic resonance imaging (fMRI). fMRI provides unique spatial information compared to ERPs. However, the few available fMRI studies have often analyzed errors post hoc rather than using tasks explicitly designed to elicit sufficient error trials, often resulting in underpowered analyses. The current study aims to address these gaps by implementing a paradigm optimized for error generation while incorporating social scrutiny. Moreover, the study examines distinctions between clinical symptomatology and naturalistically-sampled worry, in an attempt to disentangle trait-level psychopathology from momentary affect, while assessing associations among error processing, social scrutiny, and brain function in healthy youth and youth seeking treatment for anxiety disorders.

Error processing and anxiety

Erroneous behavior is linked to potential failure in control processes. Therefore, monitoring and evaluating one’s performance for mistakes facilitates cognitive control [5–7],. Prior ERP and fMRI work implicated anterior cingulate cortex and medial prefrontal cortex regions/mapping performance evaluation and adaptive control functions in error processing [8, 9]. While many psychiatric disorders show some level of cognitive control deficit, pediatric anxiety pediatric anxiety is often associated with relatively small impairments or even enhancements on cognitive control tasks involving error processing [6, 7]. However, the relation between anxiety and cognitive control is complex, with different facets of anxiety and social contexts potentially influencing control processes more specifically error processing in distinct ways [10, 11]. Specifically, the magnitude of error-related brain activity, as indexed by the error-related negativity (ERN) component of ERP, often correlates positively with levels of anxiety across many studies in different age groups in clinical and community samples [3, 12–14]. Rather than indicating a deficit in cognitive control larger ERN amplitude may reflect heightened sensitivity to negative consequences, indicating greater engagement of error processing brain regions, such as the anterior cingulate cortex, especially when mistakes carry social consequences. The ERN has been viewed as one of the best-established biomarkers of anxiety, and its rapid deployment and phasic nature suggest that it indexes a process that has been termed “reactive control”. This process involves phasic recruitment of control mechanisms in response to unexpected salient events, potentially to respond rapidly to such events [15]. While ERN findings, generated via ERPs, are relatively consistent, findings on error processing utilizing fMRI studies appear far less consistent. In fact, some studies find no differences [16], and others even find reduced engagement of error processing regions when comparing anxious and non-anxious individuals [6, 17–20]. This could reflect the differing temporal sensitivity of ERPs and fMRI, the weaker reliability of fMRI than ERP methods, or the rapid nature in which reactive control is deployed. Given these mixed findings and prior data on weak reliability [21, 22], it is critical to consider whether inconsistent findings are due to measurement error of error processing indices derived from fMRI.

Error processing, anxiety and social context

Adolescence, a critical period for the onset of anxiety disorders is characterized by heightened sensitivity to social scrutiny, specifically peer evaluation [23, 24]. For individuals with significant anxiety specifically, this developmental sensitivity may play out in cognitive processes such as error processing [25, 26]. Recent electroencephalography (EEG) studies extended research on anxiety and error processing into the realm of social scrutiny. Specifically, compared to making errors while alone, making errors when being observed by peers enhances the ERN in youth; the magnitude of this increase correlates with levels of ongoing anxiety [3]. This effect is particularly pronounced in children high in behavioral inhibition. This early childhood temperament involves high levels of wariness that manifest in reduce speech and movement when faced with novelty [3, 18]. The temperament is a risk factor for anxiety disorders [3, 18]. Such findings from EEG studies extend a wealth of fMRI research on hypersensitivity to social scrutiny in pediatric anxiety disorders [24, 27]. However, no fMRI studies on social scrutiny in pediatric anxiety utilize a cognitive control task powered to examine error processing. Moreover, even in the absence of social scrutiny manipulations, findings from fMRI studies on cognitive control in pediatric anxiety remain inconsistent [3, 28]. These inconsistencies may stem from the poor reliability of many task-based fMRI paradigms. Therefore, more fMRI research in pediatric anxiety is needed that considers both reliability and effects of social scrutiny. Differences in how anxiety symptoms are operationalized may also contribute to these discrepancies [6]. While questionnaire-based assessments may capture overall anxiety severity over longer time spans, they may not reflect the day-to-day fluctuations in worries that contribute to anxiety-related dysfunctions. Here, we capitalize on two approaches and assess anxiety via a well validated, dual-reporter questionnaire measures as well as smartphone-based naturalistic daily probes.

Understanding how adolescents process their own mistakes while under social observation has clinical implications. For example, altered error processing has been linked to risk for future anxiety, persistence of symptoms, and treatment response [29, 30]. Moreover, identifying the ways in which the brain responds to errors may help identify youth who face difficulty describing their symptoms but who struggle when their performance is being observed. This work could inform intervention targets that reduce observation-related effects.

Current study

Despite advances, inconsistent neural correlates of anxiety exist in fMRI studies. Unlike many processes, there are no robust behavioral indices of error responses, underscoring the need to address psychometric challenges of neural activation markers. The present study uses a modified fMRI flanker task to examine the neural activity associated with error processing in peer versus alone contexts. Specifically, we addressed three aims: (1) evaluate the neural activity associated with error processing and social context, (2) assess how individual differences in anxiety symptoms relate to the neural correlates of error processing and social context, and (3) how naturalistically-sampled daily worries relates to neural correlates of error processing and social context. In addition to these primary aims, we also evaluated the reliability of task-evoked neural activation, and the results are reported in Supplement 3. We expected anterior cingulate cortex and insula and dorsomedial prefrontal cortex regions to be involved in error processing with activities in social-cognition and social-evaluation region, specifically the medial prefrontal cortex and temporal parietal junction further modified by social context. Given inconsistencies in fMRI literature on anxiety-related differences in error activation we did not specify a priori hypotheses activation patterns linked to anxiety.

Methods

All procedures were approved by the National Institute of Mental Health Institutional Review Board. Consent and assent were obtained from parents and participants respectively. Recruitment involved community advertisements and monetary compensation for participation.

Participants

The study included 98 youths (M_age = 13.9, SD = 2.7, range 8–18), including 53 youth seeking treatment for an anxiety disorder. Anxiety diagnoses (at least one of the following generalized, social, separation-anxiety disorder) were established using a semi-structured clinical interview with a trained clinical psychologist (the Schedule for Affective Disorders and Schizophrenia for School-Age Children–Present and Lifetime Version-KSADS) [31], and 45 youth free of psychopathology (i.e., healthy controls). We included both treatment-seeking anxious youth and healthy controls but analyzed anxiety dimensionally to maximize power and align with contemporary transdiagnostic models. Although the task was designed to elicit sufficient error trials for most participants, a subset was excluded based on a priori error-count criteria. Of the 98 participants, 36 participants were excluded due to incomplete data (n = 6), accuracy < 70% on the task (n = 10), insufficient commission error during incongruent trails < 18 for peer and/or alone context (n = 14), and not passing fMRI quality control thresholds (n = 6). Characteristics of the 62 participants in the main analyses appear in Table 1; see supplementary material 1 for more details on excluded participants (detailed exclusion criteria can be found at https://clinicaltrials.gov/study/NCT00018057).

Table 1.

Sample characteristics

		Healthy Control (n = 30)	Anxiety (n = 32)
		N (%)	N (%)	p-values
Demographics
Sex	Female	17 (57)	21(66)	0.644
	Male	13 (43)	11 (34)
Race	White	11 (37)	22 (69)	0.008
	Black or African American	9 (30)	1 (3)
	Asian American	2 (7)	0
	Multiple Race	6 (20)	7 (22)
	Not Reported	2 (7)	2 (6)
Ethnicity	Latina or Hispanic	2 (7)	9 (28)	0.054
	Not Latina or Hispanic	27 (90)	22 (69)
	Not Reported	1 (4)	1 (3)
Combined Family Income	<$14,999	0	0	0.216
	$15,000 - $24,999	1 (3)	0
	$25,000 - $39,999	2 (7)	0
	$40,000–59,999	0	0
	$60,000–89,999	3 (10)	6 (19)
	$90,000 - $179,999	8 (27)	8 (25)
	> $180,000	9 (30)	15 (47)
	Unknown	7 (23)	3 (9)
Highest Level of Household Education	< High School	0	1 (3)
	High School	1 (3)	0
	Partial College	5 (17)	0	0.002
	4-Year College	5 (17)	6 (19)
	Graduate/professional	14 (47)	25 (78)
	Unknown	5 (17)	0
Anxiety Diagnoses	Generalized Anxiety		7 (21.9)
	Social Anxiety		21 (65.6)
	Separation Anxiety		2 (6.3)
	Specific Phobia		1 (3.1)
	Panic disorder		1 (3.1)
		Mean (SD)	Mean (SD)
	Age (y)	15.0 (2.4)	13.9 (2.1)	0.067
		Mean(SD)	Mean(SD)	p-values
Individual Differences
	IQ	110 (12.5)	115.2(12.9)	0.112
Anxiety Severity	SCAREDa Child	6.8 (6.0)	29.9 (14.6)	< 0.0001
	SCAREDa Parent	2.2 (2.9)	23.4 (11.5)	< 0.0001
	EMA b	1.2 (0.3)	1.5 (0.4)	0.006
Task Performance	Reaction Time	442.1 (136.6)	441.2 (92.6)	0.489
	Accuracy	0.84 (0.37)	0.85 (0.36)	0.009
	Incongruent Errorc	73.0 (17.5)	74.0 (17.1)	0.821

^a SCARED = Screen for Child Anxiety Related Disorders (average reported here are from total scores); ^bEMA = Ecological Momentary Assessment; ^cIncongruent Error: Average of incongruent commission errors n = 16 participants were missing EMA-worried ratings; n = 2 were missing SCARED values

fMRI task

Participants completed a modified version of the Erikson flanker task [32], in which they indicated the direction of a middle arrow (target) flanked by two arrows on either side, either facing the same (congruent condition) or opposite direction (incongruent condition). Participants performed two runs while deceived to be observed by a same age and sex peer (i.e., peer) and two runs while unobserved (i.e., alone). Order of peer-versus-alone context was counterbalanced with equal numbers of trials in each social context.

For peer runs, participants exchanged information with a fictional peer, who participants were told was observing them and would provide some prediction about their performance. All communications involved pre-recorded audio files as used in prior work [28, 33]. For alone runs, participants were told they would be performing the task alone without the peer present. All participants were debriefed. Each run comprised three blocks. After each block, participants received one of three computer-generated messages based on performance: “be more accurate” for accuracy below 75%, “respond faster” for accuracy above 90%, and “good job”: for accuracy between 75% and 90%. The participants were informed ahead of time that feedback was computer-generated, based on performance.

Whenever possible to support research on reliability, participants completed the Erikson flanker task twice, ~ 48 days apart (M = 48.14 +/- 51.26 days), the second time without the peer manipulation. Forty-six of the 62 subjects completed the task in Smith et al. [34], which comprised 432 trials of the alone context only administered across four 6-minute runs as depicted in supplementary material 1.

Clinical measures

Youth and parents reported anxiety symptoms on the 41-item Screen for Child Anxiety Related Disorders (SCARED) questionnaire (n = 60) [35] within three months of the MRI visit. An average of the total score from the child-parent reports (SCARED-composite) was calculated. Greater scores indicate more severe anxiety symptoms. The SCARED has good construct validity and reliability [35], moderate parent-child agreement in our sample (intraclass correlation coefficient = 0.66), and good internal consistency (α = 0.80).

A subset of participants (n = 46) completed at least three days of Ecological Momentary Assessment (EMA) on a smartphone within six months of the fMRI visit, assessing different emotions (details in supplement). Participants were prompted three times a day. Current analyses included participants’ average rating on the prompt: “At the time of the beep, I felt worried or scared.” which was rated on a 5-point scale (1 = Not at all, 5 = Extremely). For brevity, we refer to this measure as EMA–worried throughout the manuscript.

fMRI data acquisition and preprocessing

MRI scans were collected on two identical 3 T GE MR750 scanners with a 32-channel head coil. A sagittal structural scan was acquired as a T1-weighted, magnetization-prepared, rapid-acquisition gradient echo (MPRAGE) sequence with the following parameters: echo time (TE) = min full, inversion time (TI) = 425 ms, flip angle = 7degrees, the field of view (FOV) = 25.6 mm², and matrix = 256 × 256fx256. T*2 weighted echoplanar image volumes were acquired with X contiguous interleaved axial slices with the following parameters: TE = 25 ms, repetition time (TR) = 2000 ms, flip angle = 60 degrees, FOV = 24 mm, matrix = 96 × 96 × 96, slice thickness = 3 mm.

All imaging analyses were conducted using Analysis of Functional NeuroImages version (AFNI, version 24.1.22) [36]. We applied the standard preprocessing including: despiking, slice timing correction, non-linear registration to MNI template, warping into standard space, and spatial smoothing with a 6.5-mm full-width half maximum smoothing kernel (see supplementary material 1 for the quality control criteria used in this study). We specified to smooth to a desired blur size to assure similar smoothness is achieved across scanners. Pairs of TRs were censored if: (1) the motion shift, defined as Euclidean norm of the derivative of the translation and rotation parameters, exceeded 1 mm between TRs, and (2) more than 10% of a volume were outliers. Participants were excluded if more than 15% of data was removed due to censoring and the fraction of censored TRs was ≥ 20%. Anatomy-based image correction (ANATICOR) [37] was included for artifact reduction. At the individual level, a general linear model including six motion regressors (X, Y, Z, Pitch, Yaw, Roll) and six task condition regressors, time-locked to stimulus onset reflecting trial type (congruent, incongruent) and error condition (commissions, omissions, and correct) was specified for each social context (peer, alone). Hence, separate beta estimates were computed for each of the four trial types (incongruent error alone, incongruent correct alone, incongruent error peer, incongruent correct peer) for each participant. These condition-wise beta maps and served as inputs for our group level analysis.

Statistical analyses

Three whole brain linear mixed effects models (LME) evaluated error-specific processing (AFNI’s 3dLMEr) [38]. Only individuals with at least 18 commission errors during incongruent condition for peer and/or alone context were included in the three analyses examining error processing. Each model contrasted commission errors and correct responses in incongruent trials in the peer versus alone context. Between-subjects variables are described below. All analyses used a whole-brain gray matter mask including voxels where at least 90% of participants had signal (98,386 voxels) with a 2-sided, p <.005 threshold, with multiple testing correction of Ɑ = 0.05 using AFNI’s 3dClustSim. Smoothness and cluster-size estimation using the mixed auto correlation function (-acf flag in 3dFWHMx) generated fitted parameters for Gaussian (a, b) and mono-exponential (c) components (a = 0.5852, b = 3.3657, c = 10.3933). Cluster sizes of k > 56 were considered significant. Mean activity values for clusters were extracted using 3dROIstat for post-hoc analyses using R (v. 4.3.1) [39]. For all the models, the subject was set as the random intercept, with age and the number of commission errors as covariates. All continuous variables were grand mean-centered. See supplementary material 2 for statistical analyses of analogous group-level LME models contrasting correct responses during incongruent and congruent trials. Effect of Social Context on Error-Related Brain Activity (fMRI Model 1). To examine how social context impacts error processing, a whole-brain group-level LME model (n = 62) was performed with error contrast and social context as within-subjects variables.

BOLD ~ Condition (incongruent error, incongruent correct) * social context (alone, peer) + (1|participants) + age + number of commission errors.

Effect of Anxiety Severity on Error-Related Brain Activity (fMRI Model 2). The impact of anxiety symptom severity was examined using the SCARED-composite. A whole brain group-level LME model (n = 60) was performed with error contrast and social context as within-subjects variables and SCARED-composite as a between-subjects continuous variable.

BOLD ~ Condition (incongruent error, incongruent correct) * social context (alone, peer) * SCARED-composite + (1|participants) + age + number of commission errors.

Effect of Daily Worry on Error-Related Brain Activity (fMRI Model 3). The impact of daily worry was examined using the average rating from the EMA’s momentary worried item (EMA–worried). A whole brain group-level LME model (n = 46) was performed with error contrast and social context as a within-subjects variable and EMA–worried as a between-subjects continuous variable.

BOLD ~ Condition (incongruent error, incongruent correct) * social context (alone, peer) * EMA-worried + (1|participants) + age + number of commission errors.

Results

Significance testing for sample characteristics comparisons between anxiety and healthy control groups is reported in Table 1, with additional details provided in supplementary material 1.

Whole-brain fMRI results

Table 2 contains a summary of significant clusters.

Table 2.

Significant Associations of Neural Activity with Main Effect of Social Context and Interaction Between Social Context With SCARED-composite and EMA-worried

K	CM RL	CM AP	CM IS	Mean	SEM	t	p	Region
Main Effect of Social Context
237	−1	52	31.5	−3.29	0.03	−4.06	< 0.001	L Precuneus; L Posterior Cingulate Cortex; R Posterior Cingulate Cortex; R Middle Cingulate Cortex; R Precuneus
235	2.8	−60.5	−5.7	−3.46	0.03	4.45	< 0.001	L Mid Orbital Gyrus; L Superior Medial Gyrus; R Mid Orbital Gyrus; R Superior Medial Gyrus
183	−49.8	61	32.3	−3.45	0.03	−4.38	< 0.0001	R Angular Gyrus; R Middle Occipital Gyrus; R Inferior Parietal Lobule
133	−50.4	−5.3	−26.8	−3.24	0.03	−4.71	< 0.0001	R Medial Temporal Pole; R Middle Temporal Gyrus; R Inferior Temporal Gyrus
130	−25	7.4	−19.4	−3.47	0.04	−6.06	< 0.001	R ParaHippocampal Gyrus; R Hippocampus; R Amygdala
128	35.6	71.9	−25.2	3.44	0.04	4.06	< 0.0001	L Cerebellum (Crus_1); L Cerebellum (Crus_2)
97	−45.4	29	22	−3.46	0.06	−4.14	< 0.0001	R Rolandic Operculum; R SupraMarginal Gyrus; R Superior Temporal Gyrus; R Insula Lobe
95	−0.1	82	−16.5	3.51	0.06	4.09	< 0.0001	L Cerebellum (Crus_2); Cerebellar Vermis (7); R Cerebellum (Crus_2); Right Cerebellum (Crus_1); L Cerebellum (Crus_1)
93	−0.2	92.7	15.6	3.35	0.04	3.38	0.001	L Calcarine Gyrus; L Cuneus; R Cuneus; L Superior Occipital Gyrus; R Calcarine Gyrus
87	47.3	66.8	40.9	−3.45	0.05	−4.31	< 0.0001	L Angular Gyrus; L Middle Occipital Gyrus
86	27.7	−23.3	49.3	−3.23	0.03	−3.57	< 0.0001	L Middle Frontal Gyrus; L Superior Frontal Gyrus
84	−40.9	13.5	41.5	−3.22	0.04	−3.16	0.002	R Precentral Gyrus; R Postcentral Gyrus
68	29.8	10.5	−17.2	−3.33	0.05	−4.85	< 0.0001	L Hippocampus; L Fusiform Gyrus; L ParaHippocampal Gyrus; L Amygdala
62	44.4	16.7	38.1	−3.33	0.05	−3.59	< 0.0001	L Postcentral Gyrus; L Inferior Parietal Lobule; L SupraMarginal Gyrus
61	47.2	24.8	15.6	−3.17	0.03	- 3.99	< 0.0001	L Rolandic Operculum; L SupraMarginal Gyrus; L Superior Temporal Gyrus; L Postcentral Gyrus
Social Context X SCARED-composite
152	−58	22.6	6.6	−3.28	0.03	−5.04	< 0.0001	R Superior Temporal Gyrus; R Middle Temporal Gyrus
69	57.5	12.7	6.8	−3.40	0.05	−3.77	< 0.0001	L Superior Temporal Gyrus; L Heschls Gyrus; L Rolandic Operculum; L Middle Temporal Gyrus; L Postcentral Gyrus
57	40.4	−56.2	4.8	−3.33	0.05	−4.52	< 0.0001	L Middle Frontal Gyrus
Social Context X EMA-Worried
105	49.4	−8.4	20.4	3.45	0.05	4.57	< 0.0001	L Inferior Frontal Gyrus (p. Opercularis); L Precentral Gyrus; L Inferior Frontal Gyrus (p. Triangularis); L Rolandic Operculum
84	4.0	66.9	−1.8	3.42	0.05	3.43	0.001	Cerebellar Vermis (6); L Cerebellum (VI); L Lingual Gyrus; Cerebellar Vermis (4/5); L Cerebellum (IV-V); R Lingual Gyrus
69	46.8	67	−13.8	3.44	0.06	3.52	0.001	L Cerebellum (Crus_1); L Inferior Occipital Gyrus; L Fusiform Gyrus; L Inferior Temporal Gyrus
68	−9.6	68.3	10.9	3.14	0.03	2.77	0.006	R Calcarine Gyrus; R Lingual Gyrus; L Calcarine Gyrus; L Lingual Gyrus
60	−3.3	21.0	39.1	3.35	0.06	5.27	< 0.0001	R Middle Cingulate Cortex; L Middle Cingulate Cortex
58	12.9	88.6	24.2	3.30	0.07	2.71	0.007	L Superior Occipital Gyrus; L Cuneus; L Middle Occipital Gyrus
56	−39.5	80.2	7.1	3.29	0.06	2.11	0.036	R Middle Occipital Gyrus; R Inferior Occipital Gyrus

Anatomical locations are based on Eickhoff-Zilles macro labels from N27 (MNI space); K = voxel size of cluster. L = Left, R = Right, CM = Center Mass (Coordinates are in LPI where negative x is left, negative y is posterior, and negative z is inferior), RL =Right to Left (x), AP = Anterior to Posterior (y), IS = Inferior to Superior (z)

fMRI Model 1

While no significant error-by-social context interactions arose, 19 clusters reflected the main effect of error condition (incongruent error, incongruent correct). This was robust in several regions, including the bilateral anterior cingulate cortex, bilateral posterior cingulate cortex, and bilateral precuneus (see supplementary material 1 for a full list with statistical values). Additionally, 15 clusters reflected the main effect of social context (peer, alone), with decreased activity for the peer relative to alone context in bilaterial precuneus (t(244) = −4.06, p <.001; Fig. 1a), bilateral mid orbital gyrus (t(244) = −4.45, p <.001), right parahippocampal gyrus/amygdala, (t(244) = −6.05, p <.001; Fig. 1b), and right rolandic operculum (t(244) = −2.14, p <.001).

Fig. 1.

Neural responses to social context. Shown are selected significant clusters and their associated plot. (a) Bilateral precuneus and (b) right parahippocampal gyrus showing the main effect of social context. Suprathreshold results are highlighted as opaque and outlined; sub-threshold regions are visible with decreasing opacity as the voxel’s statistical significance decreases

fMRI Model 2

While no error-by-social context-by-SCARED-composite interaction arose, three clusters reflected a significant social context-by-SCARED-composite interaction. Higher SCARED-composite related to decreased activity during the peer compared to alone context in the bilateral right and left superior/middle temporal gyrus, (right: t(236) = −5.04, p <.0001; Fig. 2a; left: t(236) = −3.77, p <.0001), and left middle frontal gyrus, (t(236) = −4.52, p <.0001; Fig. 2b).

Fig. 2.

Impact of Individual differences (SCARED-composite and EMA-worried) on neural responses to social context. Shown are selected significant clusters and their associated regression plots. (a) Left superior temporal gyrus and (b) left middle frontal gyrus activity showing the impact of SCARED-composite on peer context. (c) Left inferior frontal gyrus and (d) bilateral middle cingulate cortex activity showing the impact of EMA-worried on social context. Suprathreshold results are highlighted as opaque and outlined; sub-threshold regions are visible with decreasing opacity as the voxel’s statistical significance decreases

fMRI Model 3

While no error-by-social context-by-EMA-worried interaction arose, four clusters reflected a significant social context-by-EMA-worried interaction. Higher worry related to increased activity during the peer compared to alone context in the left inferior frontal gyrus (t(178) = 4.56, p <.0001; Fig. 2c) and bilateral middle cingulate cortex (t(178) = 5.27, p <.0001; Fig. 2d).

Two-way interactions are presented in the Supplementary 1, S3.1 Adjusted Interaction Models.

Discussion

Two key findings emerged from this study. First, across all participants, making an error (i.e., main effect of error condition) was associated with increased activation in a broader regions involving performance monitoring and cognitive control, including the anterior cingulate cortex/supplementary motor area, anterior insular/inferior frontal and parietal areas. This pattern aligns with prior work indicating that error processing reliably engages a cingulate-opercular (i.e., neural alarm system) centered on the anterior cingulate cortex and anterior insula, which provides further support for the validity and reliability of task as probe for error processing [40, 41]. Additionally, performing the task while being observed by a peer (i.e., main effects of social context) was associated with relative deactivation in precuneus, inferior parietal, superior, and middle temporal regions, as well as the hippocampus, some of which lie within the default mode network (DMN). These deactivations may indicate that peer observation may engage more processing resources in the service of supporting task-relevant operations [42] compared to the alone condition. Consistent with these findings, prior research suggests that increases in task demands potentiate DMN deactivation.

Of note, these findings pertained only to main effects; their interaction (i.e., error condition and social context) did not reach cluster-corrected significance. Hence, the presence of a peer did not measurably impact brain function in areas modulated by error processing. Our data do not allow us to draw conclusions as to why we did not find this effect. However, it is of note that some prior work in a community sample has shown that the magnitude of error-related signal was predicted by the motivational significance of the performance (i.e. competitive or cooperative task settings) and prior work in treatment-seeking anxious youth did not find associations between error processing and anxiety levels in youth that were not also characterized by temperamental risk factors [34, 43]. Alternatively, null findings could be due to either (1) limited power of voxel wise tests of interactions, particularly higher-order interactions or (2) our performance-based exclusion criteria (i.e., sufficient commission errors) which likely improved the reliability of the error contrast, however, narrowed between-subject variance.

Second, we found that youth with higher levels of anxiety and worry showed stronger neural sensitivity to social context (i.e., interactions between social context and anxiety). In Model 2, higher anxiety symptoms (SCARED-composite) were associated with a larger activation difference between alone and peer-observation context. Youth with more anxiety symptoms showed increased activation in superior and middle temporal and middle frontal gyri in the alone relative to peer-observation context, potentially reflecting altered attentional engagement under perceived peer scrutiny. In Model 3, higher levels of daily worry (i.e., EMA-worried) were link to greater activation in regions largely outside of networks relevant to task processing including in the inferior frontal gyrus and middle cingulate cortex during peer-observation compared to the alone context. Together these finding indicate that both overall anxiety symptoms and momentary worry are related to how strongly social context impact neural responses during task performance, however, they do so in some distinct brain regions and directionality.

We failed to detect any three-way interactions involving error condition, social context and the between-subject measures of anxiety and momentary worry. In other words, we found no evidence that anxiety or daily worry altered error-related activation, either directly or through social context. Rather, our findings suggest that anxiety and momentary worry may influence broader, albeit distinct, evaluative and attentional networks rather than specifically error-related processes. Anxiety symptoms were linked to regions associated with social processing and attention. These included bilateral superior and middle temporal gyri as well as left middle frontal gyrus, potentially reflecting altered attentional engagement under perceived peer scrutiny. In contrast, momentary worry, assessed via EMA, related to activation in regions largely outside of networks relevant to task processing, including inferior frontal gyrus, and middle cingulate cortex. These patterns may reflect shifts in control dynamics during peer observation, with anxiety and momentary worry uniquely influencing cognitive resource allocation. Anxiety symptoms and levels of momentary worry are only partially related constructs. Anxiety symptoms were measured using a questionnaire, while worry was measured using EMA, which is designed to eliminate recall biases contained in questionnaires [44]. Agreement among these indices is moderate to poor [45, 46]. Moreover, each of the two anxiety-relevant constructs related to different brain regions with unique functions. Thus, distinct findings for the two anxiety measures may reflect variations in construct features or measurement properties as well as psychological functions and their associated neural architecture.

A few features of our study design may enhance sensitivity to anxiety-related neural correlates. First, it may be important to engage youth in demanding tasks, such as those that require inhibitory control, to reveal the neural correlates of anxiety. This could in turn relate to many factors. For example, demanding tasks might have a unique psychological impact on youth with and without anxiety, evoking a state of anxiety related to performance on the task. Second, psychometric factors could account for enhanced sensitivity. Supplementary material 3 includes analyses examining reliability of the relevant events for incongruent error vs. incongruent correct, a focus of our analyses, contrast which showed fair-good between session reliability by task-fMRI conventions (i.e., ICC indexing the similarity of BOLD contrast magnitude across two time points) [47]. Enhanced reliability facilitates sensitivity to individual differences. In a related fashion, we utilized a task that contained many replicates of each trial type, and we restricted analyses to subjects with enough errors to generate reliable data. Future work might consider these and other possible explanations for specific associations with incongruent events.

Limitations

These results should be interpreted considering several limitations. First, the modified fMRI flanker task used a simulated peer experience; authentic interactions may magnify individual differences. However, most prior studies also utilize simulated peer experiences, given the difficulties of implementing such authentic experiences in the scanner. Hence, this limitation does facilitate comparisons with prior results. Second, while test-retest reliability of the error contrast for the Flanker task was strong, we could not assess reliability of the social manipulation, as ethical constraints required that subjects be debriefed. It is difficult on both ethical and procedural grounds to evaluate test-retest reliability in studies utilizing deception. Third, due to our modest sample size and limited variance we only controlled for age and number of errors made. Fourth, the relatively affluent composition of the sample limits generalizability, and the sample size is not large. Given the complexity and associated cost with the procedures, demonstrating interesting findings in relatively small, non-representative samples provides justification for implementing such work in the future.

Future directions

Future research might investigate effects of other social manipulations in larger, more diverse samples. Such future work might examine how other contextual factors, such as task difficulty or social familiarity, influence neural patterns. Future work might also examine the temporal dynamics of error-related neural responses in social context using integrated methods with improved temporal resolution. Finally, studies might use longitudinal or treatment-related designs to model developmental trajectories or relations with treatment response during adolescence.

Conclusion

Findings highlight the interplay among neural correlates of conflict processing, social context, and individual differences during adolescence, a critical developmental period for self-regulation and social cognition. Social context during conflict processing interacted with anxiety symptoms and worry ratings in distinct ways. Increased anxiety symptoms were related to increased brain activation in one set of brain regions, whereas increased daily worry was related to a different activation in another set of brain regions. These divergent relations among brain function, social context, and anxiety suggest that dissociable underlying mechanisms may relate to distinct facets of anxiety. Taken together, the results emphasize the critical role of social context and individual differences in anxiety and worry in the social context of conflict processing, particularly during adolescence.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

PK: Conceptualization; Data curation; Formal analysis; Methodology; Project Administration; Validation; Visualization; Investigation; Writing – original draft; Writing – review & editing. AP: Data curation; Methodology; Project Administration; Investigation; Writing – review & editing. EPM, JLB, OS, and MM: Project Administration; Investigation; Writing – review & editing. EC: Methodology; Project Administration; Investigation; Writing – review & editing. KK: Conceptualization; Data curation; Methodology; Validation; Writing – review & editing. MB: Conceptualization; Data curation; Methodology; Validation; Writing – review & editing. SPH: Supervision; Conceptualization; Supervision; Data curation; Formal analysis; Methodology; Project Administration; Validation; Visualization; Writing – original draft; Writing – review & editing. DSP: Supervision; Conceptualization; Funding acquisition; Methodology; Project administration; Resources; Software; Writing – original draft; Writing – review & editing.

Funding

This study was funded by the National Institute on Mental Health (NIMH) Intramural Research Program, Principal Investigator [PI]: Daniel S. Pine (ZIA-MH002781 and ZIA-MH002782).

Data availability

.In compliance with current NIH data-sharing policies, data from participants who have provided consent for their data to be shared in a public repository are available on OpenNeuro (doi: 10.18112/openneuro.ds007179.v1.0.0). The analysis code can be found on GitHub (https://github.com/NIMH-SDAN/Neural-Response_to_Error_Social-Context).

Declarations

Ethics approval

All procedures of this study were approved by National Institute of Mental Health Institutional Review Board and certified that the study was performed in accordance with ethical standards as laid down in the 1964 Declaration of Helsinki. Consent and assent were obtained from parents and participants respectively. Recruitment involved community advertisements and monetary compensation for participation.

Competing interests

The authors declare no competing interests.