Neural and Behavioral Measures of Stress-induced Impairment in Error Awareness and Post-error Adjustment

Abstract

Exposure to stress negatively affects error processing, but the impact of stress on error awareness remains to be determined. In the present study, we examined the temporal dynamics of error awareness and post-error adjustment following acute stress. Forty-nine healthy men were randomly assigned to the control (n = 26) or stress group (n = 23). After stress induction, participants completed the error awareness task, and their brain activity was assessed by electroencephalography. Compared to the control group, the stress group demonstrated lower error awareness accuracy and smaller Pe (error positivity) and ΔPe amplitudes following aware error responses, which indicated impairment of error awareness following stress. Furthermore, the stress group had lower accuracy in post-aware error responses than in post-unaware error responses and the control group, which indicated poor post-error adjustment following stress. Our results showed a stress effect on sequential stages of error processing. Stress induces impaired error identification, which further generates maladaptive post-error performance.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12264-023-01154-2.

Introduction

Stressors are risk factors for our physical function and cognitive processing. The stress response modulated by the activity of the sympathetic-adrenal-medullary (SAM) and hypothalamic-pituitary-adrenal (HPA) axes is the basis for stress effects [1, 2]. When exposed to stress, the SAM axis is activated rapidly. The stress level of catecholamines and dopamine from the SAM leads to "fight or flight" responses in individuals. Moreover, cortisol produced from the HPA axis helps mobilize bodily resources in regaining balance from stressors [3]. These stress hormones always up-regulate amygdala-mediated emotional responses but impair higher prefrontal cortex (PFC) function through the glucocorticoid or mineralocorticoid receptors expressed in these brain regions [4, 5]. Accumulating evidence suggests that acute stress impairs executive functions [6–8].

Error processing, which involves error monitoring and post-error adjustment, is one of the essential functions of top-down cognitive control [9]. The typical phenomenon of error processing is post-error slowing (PES), indicated by a longer reaction time after erroneous action [10, 11]. It is assumed that error responses induce a more conservative response strategy on the speed-accuracy trade-off curve in individuals, which is mainly aimed at avoiding further error commission [12, 13]. Accordingly, stress may lead to adverse effects on error-related regulation. Evidence has indicated impaired error processing following stress. Research by Clemans and colleagues (2012) revealed impaired error monitoring in patients with post-traumatic stress disorder [14]. Other studies have found adverse stress effects on error-monitoring systems [15–17]. Furthermore, a few studies have reported impaired post-error behavior following impaired error monitoring under stress [18], indicating that the stress effect on error processing might be full-stage, including error monitoring and post-error adjustment stages. However, research on the stress effect on the different stages of error processing has yet to reach a consensus.

Notably, error responses sometimes remain undetected, and the post-error adjustment is only initiated as individuals recognize an error signal [19–21]. The conscious perception of error (error awareness) subserves the execution of post-error adjustment. Previous studies have yet to reveal the function of error awareness and its regulation of post-error processing following stress. Moreover, accumulating evidence suggests that error awareness is generated from different sources, such as posterior medial frontal cortex (pMFC) activity, unexpected sensory input, and changes in the autonomic nervous system [22]. Brain imaging studies have shown that the dorsolateral PFC [23], insula cortex, anterior cingulate cortex (ACC), submarginal gyrus, and supplementary motor area [24, 25] are associated with the generation of error awareness; these regions are vulnerable to stress and stress-induced hormones [26–29]. An adapted Go/No-Go task that requires participants to indicate error awareness via a keypress is often applied to explore error awareness [30–32], and the explicit conscious response is used in the estimation of the level of error awareness following stress.

Prior research has identified two electroencephalogram (EEG) components that indicate error awareness: error-related negativity (ERN) and error positivity (Pe). The ERN is a negative event-related potential (ERP) that peaks within 100 ms following an error response. ERN is always recorded over the fronto-central region [33, 34] and indicates early/automatic error detection. The Pe is a later positive deflection recorded over centroparietal regions within 200–500 ms after an error response [35]. The Pe indicates awareness, motivational or emotional assessment, or evident accumulation of error commission [36, 37]. Some studies have suggested that the ERN is sensitive to error perception [38], but the Pe component is generally recognized as representing the gathering of evidence that generates error awareness [39, 40]. Neuroimaging and source localization studies showed that the ACC is the source of the ERN. The Pe arises from the activity of a network of regions: the rostral ACC, bilateral inferior parietal, and bilateral midfrontal cortices [41–43]. These two electrophysiological indicators are essential in revealing the neural basis of activity under error awareness following stress. This experiment investigated the components of ERN and Pe to reveal the temporal dynamic modulation of error awareness following acute stress.

Combining the Trier Social Stress Test (TSST) [44] and the Error Awareness Task (EAT) with the ERP technique, this study aimed to reveal the temporal dynamics of error awareness and its impact on post-error adjustment following acute stress. The physical and mental states of participants were measured by collecting salivary cortisol, heart rates, self-reported stress perception, and negative affect to determine whether this experiment successfully induced acute stress. In the EAT, participants had to identify their erroneous responses with an explicit button press as accurately as possible. The accuracy of error awareness responses and the related ERP components (ERN and Pe) following error responses were used to measure error monitoring following acute stress and normal states. Reaction times (RTs) and the accuracy of trials following error responses were applied to explore post-error adjustment. This experiment revealed the impact of stress on error monitoring and post-error regulation by comparing the ERN, Pe, awareness of error response, and behavioral adjustment after errors between the stress and control groups. We expected that acute stress would decrease the level of error awareness in the participants and manifest as changes in error-related components. Error awareness would modulate the post-error adjustment process, which assumes that impaired error recognition following acute stress would impair post-error processing. This experiment would disclose the features of multiple stages for error processing following stress.

Materials and Methods

Participants

Since previous research has revealed different stress responses and cognitive processing in different genders and during menstrual cycles [45, 46], this study involved the recruitment of 54 male college students from Southwest University (China) through online advertisements and phone interviews. Five participants were removed from further analysis due to undetectable cortisol samples (>2 cortisol samples per participant were required), few error responses (<6), or too many artifacts in EEG records. Finally, the stress induction, behavioral, and electrophysiological analyses included 49 participants (mean ± SD: 20.35 ± 2.15 years old). The stress group consisted of 23 participants, while the control group had 26 participants. All participants had yet to undergo the TSST or EAT. All participants were right-handed nonsmokers with no history of mental or physical illness. None of the participants had taken any medication in the previous week. Before the telephone interview, all participants completed the Beck Depression Inventory (BDI) [47] and the Life Events Scale (LES) [48]. Because of the effect of depression or chronic stress on the stress responses, only participants with a BDI score <8 (3.43 ± 2.25) and an LES score <20 (9.31 ± 6.38) were included in this experiment. The mean body mass index (BMI) was 21.31 ± 2.59. The details of the characteristics of the participants are listed in Table 1. The analysis for BMI [t₍₄₇₎ = −0.49, P = 0.63], BDI [t₍₄₇₎ = 0.62, P = 0.54], and LES [t₍₄₇₎ = 0.73, P = 0.47] revealed no significant between-group difference. No significant between-group difference in age was revealed (χ² = 3.48, P = 0.06). Moreover, the analysis for the state [t₍₄₇₎ = −0.42, P = 0.68] and trait anxiety [t₍₄₇₎ = −0.04, P = 0.97] revealed no significant between-group differences. In addition, all qualified participants refrained from engaging in physical activity the day before the experiment. Furthermore, participants refrained from eating within 2 h before the experiment. Before participating, all individuals provided informed consent and were given 60 Chinese Yuan. The Southwest University Human Ethics Committee approved the present study for Human Research.

Table 1.

Demographic and psychological characteristics of participants.

	Stress Group (n = 23)	Control Group (n = 26)
Age	19.74 ± 1.39	20.90 ± 2.55
BMI	22.01 ± 4.55	21.48 ± 3.00
LES	8.65 ± 6.85	9.88 ± 6.02
BDI	3.22 ± 2.28	3.62 ± 2.25
State-anxiety	36.61 ± 5.36	35.96 ± 5.50
Trait-anxiety	42.91 ± 5.50	42.85 ± 7.56

Data are the mean ± SD. BMI, body mass index; LES, Life Events Scale total score; BDI, Beck Depression Inventory total score.

Experimental Procedure

The whole procedure lasted ~2 h (see Fig. 1A). This experiment was conducted between 13:00 and 19:00. The cortisol levels of the participants were consistently low and stable during this time. All participants were randomly assigned to undergo either control or stress conditions in the experiment. Upon arrival at the laboratory, participants were fitted with EEG and electrocardiogram (ECG) acquisition devices. In addition, participants completed the State-Trait Anxiety Inventory [49]. Afterward, initial samples of saliva cortisol and heart rate were collected, and participants completed the Perceived Stress Assessment Scale and Positive and Negative Affect Scale [50] for the first time (T1). The Perceived Stress Assessment Scale measures stress levels based on a 1–10 Likert scale. Higher scores indicate higher stress levels. Afterward, the participants proceeded to practice blocks of the EAT. Following this, the stress group performed the TSST, while the control group completed a modified version of the TSST. After the stress/control task, the second set of samples for stress indicators was collected (T2). The third set of stress indexes was collected after a 10-min break to allow for peak stress-level cortisol concentration (T3). Next, all participants engaged in the formal experimental task while sitting 60 cm away from a 17-inch monitor with a refresh rate of 85 Hz and a resolution of 1024 × 768 in a room designed to reduce sound. After the EAT, the fourth set of stress index samples was collected (T4).

Fig. 1.

Experimental procedure. A Overview of the study procedure. T2 (time = 0) refers to the end of the stress test. B Procedure of the error awareness task. TSST, trier social stress test; EAT, error awareness task.

Error Awareness Task

We evaluated error awareness using an error awareness task that involved responding to a series of single-color words on a screen. The details of the error awareness task can be found in Fig. 1B. Stimulus materials were presented in six colors: white, purple, blue, yellow, green, and red. Each color was also paired with a corresponding Chinese character. The color words were as follows: white (255, 255, 255), purple (110, 50, 160), blue (0, 0, 255), yellow (255, 255, 0), green (0, 255, 0), and red (255, 0, 0). During training, all participants were instructed to follow three rules: (1) if the font color of the color word was different from the meaning of the word, they were to press the button (referred to as "go trials"); (2) if the font color of the color word matched the meaning of the word, they were to withhold the press response (known as "color no-go trials"); and (3) if the meaning of the current color word was the same as the word in the previous trial, they were also to withhold the response (known as "repeat no-go trials"). The task varied the strength of stimulus-response relationships to induce a suitable number of unaware error responses. Specifically, since the representations of the task rules competitively suppressed each other, the more prepotent rule tended to suppress the weaker rule, resulting in many errors. Furthermore, participants may ignore part of the error responses as they prioritize the prepotent rule [31]. To signal error awareness, all participants had to press a 'conscious button' during a trial immediately after an erroneous response (pressed in both no-go conditions). In the following go trial, participants had to disregard the color and meaning of the word and instead provide an aware error response by pressing a different button. Each participant completed two practice blocks of 20 trials before the formal experimental sessions. The participants performed the formal experiment after becoming familiar with the task rules.

The formal experiment presented six blocks of 210 trials, including 30 no-go and 180 go trials. The number of the repeat and color no-go trials in these six blocks was equal. Within each block, the three kinds of trials were presented pseudorandomly. The color words were presented for 800 ms and disappeared after pressing the button, and the trial interval was 700 ms. During the Go/No-Go task, half of the participants pressed the letter “A” to complete it and the letter “L” to indicate error awareness. Participants were given key assignments that were counterbalanced. Since the button press to signify error awareness would affect subsequent processing, all no-go trials were followed by at least four go trials.

Stress Manipulation and Validation

The stress group completed the TSST, which consisted of three sections: 5 min of preparation, 5 min of speech, and 5 min of mental arithmetic. In the preparation section, the participants were informed about the stress tests. In the speech section, participants completed a job interview and tried to impress the two people acting as interviewers to get the job. In the mental arithmetic section, participants were instructed to subtract 17 from 2043 and continue doing so as quickly and accurately as possible. If a wrong number was given, they had to start again from the beginning. During the stress test, the participants were monitored by two experimenters and videotaped. Finally, the participants were notified of their unsatisfactory performance and were required to redo the task. Moreover, the control group participants completed an easier version of the TSST, which had a similar time frame but entailed more straightforward tasks. These participants shared a movie, book, or recent vacation without supervision and counted forward from 0 at their own pace in increments of 15.

We investigated stress responses using various measures such as salivary cortisol, heart rate, perceived stress level, and positive and negative affect. Saliva samples were collected using a saliva collector (Salivette, SARSTEDT, Nümbrecht, Germany) and stored at −20°C until analysis. The cortisol concentration was measured using the Enzyme-linked Immunosorbent Assay [51]. A Spirit-10 wireless telemetry biofeedback instrument recorded the heart rate. ECG electrodes were placed on the chests of participants, and the data were analyzed using BioTrace+ (Mind Media, B.V., Roermond-Herten, Netherlands). Heart rate collections were taken while salivary cortisol sampling was performed. Each heart rate sampling session lasted ~3 min, and data were continuously collected during the TSST/Control-TSST (2 min sampling at T3).

EEG Data Collection

A 64-channel Brain Products system was used to record EEGs, with a standard EEG cap based on the extended 10–20 system. For data collection, the signals were recorded at a sampling rate of 500 Hz. The online reference electrode for all sites was FCz, and the average bilateral mastoids (TP9 and TP10) were used as the reference electrode for offline processing. The electrode of the horizontal electrooculogram (EOG) was placed lateral to the right eye. The electrode of the vertical EOG was positioned below the left eye. The impedances of all channels remained under 5 kΩ. The EEG data were analyzed offline using Brain Vision Analyzer 2.1 software (Brain Products, Gilching, Germany) and MatLab (MathWorks, Natick, USA). The raw data were digitally filtered using a high‐pass filter of 0.5 Hz and a low‐pass filter of 30 Hz at 24 dB/octave. The effects of eye movements and blinks were corrected using independent component analysis, and any trial with amplitudes exceeding ±80 µV was removed. After this artifact rejection, 97.61% ± 8.68% of the trials with the EAT remained without significant differences among any conditions (P >0.22).

Segments locked to the onset of error and correct responses (from −200 to + 600 ms) were extracted. After the segment, baseline corrections were applied via an interval from −200 to −100 ms. The difference waveforms between the error and correct trials were computed to select the time windows and regions of interest for the ERN and Pe. The time windows and electrode locations for error-related components were maximally determined in the grand-averaged waveforms and topography across all participants and conditions. The mean amplitudes of the ERN and Pe components were measured to indicate the activity of error monitoring. The time window for the ERN component was −20 to 30 ms relative to error response onset, and the region of interest was the fronto‐central region [(FC1 + FCz + C1 + Cz)/4]. The Pe amplitude was defined in the time window from 250 to 500 ms relative to the error response onset at the centro‐parietal region [(CP1 + CPz + CP2 + Pz + P1 + P2)/6]. Regarding correct responses, the time windows and regions of interest for correct-related negativity (CRN) and correct positivity (Pc) were identified similarly during the correct go trials. The ΔERN and ΔPe were calculated by subtracting the amplitude of the correct trials from the amplitude of the error trials, thereby isolating the ERP components related to error awareness to explain the specific cognitive functions [52].

Statistical Analyses

Stress Induction

The effects of stress manipulation on the stress indicators were analyzed via repeated-measures analysis of variance (ANOVA) with the factors Time (T1–T4) × Group (control group vs stress group). In addition, the effect of Group on BMI, BDI, LES, state anxiety, and trait anxiety was analyzed using independent sample t-tests. Age was compared between groups using a χ² test.

Go/No-Go Task

Independent samples t-tests assessed the stress effects on the accuracy and RTs during the go trials. In addition, the stress effect on the accuracy and RTs during different no-go trials, a repeated-measures ANOVA with the factors No-go Type (color no-go vs repeat no-go) × Group (control group vs stress group) was applied.

Error Awareness Task

The level of error awareness was calculated by dividing the number of aware error trials by the total number of error trials. Between-group comparisons of the error awareness rates were conducted through independent samples t-tests. Further, between-group comparisons of accuracy and RTs during both no-go trials were examined through repeated-measures ANOVAs with the factors No-go Type (color no-go vs repeat no-go) × Group (control group vs stress group).

Errors and Post-error Responses

A repeated-measures ANOVA comprising the factors Trial Type (aware error vs unaware error) and Group (control group vs stress group) for the RTs in error responses during different awareness levels was performed. In addition, the stress impact on RTs in error responses at different no-go responses was examined using a No-go Type (color no-go vs repeat no-go) × Group (control group vs stress group) ANOVA. The experiment measured the accuracy of post-error adjustment by calculating the difference between the trials that followed correct no-go trials and those that followed error no-go trials. The computational formula of accuracy is ACC_{post-error adjustment} = ACC_{post-error trial} – ACC_{post-correct trial}. To exclude the interference of global performance shifts on the measurement of performance after error responses [53], the post-error adjustment of RTs was indicated by subtracting the RTs of the pre-error trials from those of the post-error trials. The computational formula for RT was RT_{post-error adjustment} = RT_{post-error trial} – RT_{pre-error trial}. The stress effect on the post-error adjustment during different awareness levels in the first and second post-error trials was analyzed by Error Type (post-aware error vs post-unaware error) × Group (control group vs stress group) ANOVAs. When the participant detected the error response and pressed the consciousness button, the first post-error trial was the first trial following the error awareness response. If the participant failed to detect an error response, the go-trial after the error response was the first post-error trial. The second post-error trial followed the first post-error trial.

ERP Data

The group comparison in the ERN and CRN amplitudes was submitted to a Trial Type (aware error vs unaware error vs correct) × Group (control group vs stress group) ANOVA. The stress effect on the Pe and Pc was examined similarly. Furthermore, repeated-measures ANOVA with the factors Error Type (aware error vs unaware error) × Group (control group vs stress group) was used to explore the stress effects on the ∆ERN and ∆Pe.

The statistical significance level (α) was set to 0.05 (two-tailed). The Bonferroni correction was applied to correct alpha levels. Greenhouse-Geisser corrections were implemented whenever needed. Once a significant interaction was identified, the least significant difference test was applied. The effect size of significant results was indicated by part η squared (η_p²).

Under a given significance level of 0.05, with a sample size of 49 participants and a population correlation of 0.60 in the repeated measures analysis, the two‐way interaction effect of Error Type (post-aware error vs post-unaware error) × Group (control group vs stress group) on post-error adjustment indicated a medium effect (η_p² = 0.25) at a probability (1 − β) = 0.93. For ERN/CRN and Pe/Pc, the interaction of Error Type (aware error vs unaware error vs correct) × Group (control group vs stress group) detected a medium effect at a probability of 0.97. The interaction of Error Type (aware error vs unaware error) × Group (control group vs stress group) on the ∆ERN and ∆Pe identified a medium effect at a probability of 0.93.

Results

Stress Data Results

Salivary Cortisol

The interassay coefficient of variation was <4.60% in four plates (high-concentration standard samples: 0.76%, low-concentration standard samples: 4.60%). The Time × Group ANOVA indicated a significant main effect of Time, F_{(2.38, 111.94)} = 8.03, P <0.001, η_p² = 0.15. The Time × Group interaction was significant, F_{(2.38, 111.94)} = 4.49, P = 0.005, η_p² = 0.087 (see Fig. 2A). The results of the post hoc tests revealed that the stress group had significantly higher cortisol levels than the control group during T3 (P = 0.018). However, no significant between-group differences were found at other time points (P >0.48). A significant main effect of Group was not found (P = 0.28).

Fig. 2.

Results of two-way ANOVA of stress responses over time in the control (n = 26) and stress group (n = 23) (mean ± SEM). The stress group shows increased salivary cortisol (A) and heart rate (B), perceived stress level (C), and negative affect (D) compared to baseline and the control group, indicating successful stress induction. *P <0.05, **P <0.01, ***P <0.001.

Heart Rate

The results revealed a significant main effect of Time, F_{(1.87, 87.92)} = 74.97, P <0.001, η_p² = 0.62 (see Fig. 2B). The main effect of Group was significant, F_{(1, 47)} = 7.53, P = 0.009, η_p² = 0.14. The interaction of Time × Group was also significant, F_{(1.87, 87.92)} = 50.54, P <0.001, η_p² = 0.52. The simple effect tests showed that the stress group had significantly higher heart rates than the control group at T2 (P <0.001). No significant between-group differences were found during the other time points (P >0.09).

Perceived Stress Level

The ANOVA results yielded a significant main effect of Time, F_{(2.59, 121.80)} = 12.44, P <0.001, η_p² = 0.21, and a marginally significant main effect of Group, F_{(1, 47)} = 3.85, P = 0.056, η_p² = 0.08 (see Fig. 2C). Furthermore, the Group × Time interaction was also significant, F_{(2.59, 121.80)} = 11.41, P <0.001, η_p² = 0.20. During T2 and T3, the stress group reported higher perceived stress levels than the control group (P <0.001 and P = 0.002, respectively). No between-group differences occurred during the other time points (P >0.26).

Positive and Negative Affect

Regarding the negative affect, the Group × Time ANOVA revealed a significant main effect of Time, F_{(3, 141)} = 8.36, P <0.001, η_p² = 0.15. The main effect of Group was also significant, F_{(1, 47)} = 5.61, P = 0.02, η_p² = 0.11 (see Fig. 2D). The interaction of Group × Time was significant, F_{(3, 141)} = 13.28, P <0.001, η_p² = 0.22. After post hoc tests, it was found that the stress group had a significantly higher level of negative affect than the control group at T2 (P <0.001) and T3 (P = 0.03). No significant difference was found at the other time points (P >0.62). For the positive affect, the main effect of Time was significant, F_{(3, 141)} = 40.76, P <0.001, η_p² = 0.46. The Group × Time interaction was significant, F_{(3, 141)} = 2.97, P = 0.050, η_p² = 0.06. Throughout the experiment, all participants experienced a decrease in positive feelings. However, there was no significant difference between the two groups at any time point. The main effect of Group was not significant (P = 0.57).

Area Under the Curve with Respect to the Increase (AUCi)

The AUCi in salivary cortisol, heart rate, and self-report measures at each time point were calculated to indicate the stress response. The t-test was applied to explore the group difference in the AUCi. The results showed significant group differences in the AUCi of the salivary cortisol, t₍₄₇₎ = −2.62, P = 0.012, d = −0.05; heart rate, t₍₄₇₎ = −6.74, P <0.001, d = −0.85; and perceived stress level, t₍₄₇₎ = −3.82, P <0.001, d = −0.18; and negative affect, t₍₄₇₎ = −2.67, P = 0.01, d = −0.36. These results suggested that the stress manipulation was successful in the stress group.

Behavioral Results

Go/No-Go Task

The results of accuracy and RTs during different no-go trials are presented in Table 2. The independent samples t-tests for accuracy and RT of the go trial revealed no significant between-group difference (P >0.09). After the No-go Type × Group ANOVA of the accuracy during different no-go trials, a significant main effect of No-go Type was present, F_{(1, 47)} = 90.47, P <0.001, η_p² = 0.66, with higher accuracy in the repeat no-go trial (51.28% ± 21.97%) than in the color no-go trial (31.38% ± 20.20%). The main effect of the Group was significant, F_{(1, 47)} = 4.58, P = 0.038, η_p² = 0.09, and the control group had a significantly higher accuracy (46.82% ± 22.10%) than the stress group (35.13% ± 23.20%). The interaction of No-go Type × Group was not significant (P = 0.22).

Table 2.

Accuracy and RTs of the go and no-go trials.

	Trial type	Stress group (n = 23)		Control group (n = 26)
		Mean	SD	Mean	SD
Accuracy (%)	Go	96.53	2.5	97.59	1.5
	Color no-go	26.52	19.94	35.68	19.82
	Repeat no-go	43.70	23.45	57.97	18.57
RT (ms)	Go	451.66	101.78	473.77	64.96
	Color no-go	450.34	105.48	485.21	78.68
	Repeat no-go	473.86	123.53	497.92	94.55

RT, reaction time.

Error Awareness

The results of accuracy and RTs during error awareness are illustrated in Fig. 3 The t-test yielded a significant between-group difference during the accuracy of error awareness, t₍₄₇₎ = 1.97, P = 0.055, d = 0.56, with higher accuracy in the control group (65.48% ± 16.15%) than in the stress group (54.34% ± 23.31%). The between-group difference during RTs of error awareness was also significant, t₍₄₇₎ = 2.14, P = 0.038, d = 0.62, indicating faster RTs in the stress group (244.84 ± 59.03 ms) than in the control group (289.45 ± 83.37 ms). Regarding the level of error awareness in different no-go trials, the No-go Type × Group ANOVA showed a significant main effect of Trial Type, F_{(1, 47)} = 12.34, P <0.001, η_p² = 0.21, indicating that the accuracy of error awareness during the repeat no-go trial (53.88% ± 24.66%) was significantly lower than that during the color no-go trial (64.60% ± 22.25%). The main effect of Group was also significant, F_{(1, 47)} = 4.14, P = 0.048, η_p² = 0.081, showing higher accuracy in the control group than in the stress group. The Trial Type × Group interaction was not significant (P = 0.90). All effects were not significant for RTs of error awareness during different no-go trials (P >0.11). These results revealed that the task difficulty of the repeat no-go trial and color no-go trial differed, while the no-go type effect did not affect group differences.

Fig. 3.

Results of ANOVA of accuracy and RTs during error awareness in the control (n = 26) and stress group (n = 23) (mean ± SEM). A Faster reaction and lower accuracy of error awareness in the stress group. B Error awareness during the repeat no-go is lower than during the color no-go trial in both groups. *P <0.05. RT, reaction time.

To explore whether the trade-off between speed and accuracy affects the error awareness responses in the stress responders, we calculated the following indicators in the stress group and the control group: (1) the probability of recognizing error responses (P_awareness); (2) the probability of missing error responses (P_unawareness); (3) the probability of recognizing an error response after a correct response (P_{false awareness}); and (4) the probability of recognizing a correct response after a correct response (P_{correct judgment}). The t-test revealed that the P_awareness in the control group was significantly larger than that in the stress group, t₍₄₇₎ = 1.74, P = 0.055, d = 0.56. In addition, the stress group had a larger P_unawareness than the control group, t₍₅₀₎ = −2.14, P = 0.037, d = −0.59. No significant between-group difference was found for P_{false awareness} and P_{correct judgment}. The results indicated that the stress group never had more awareness responses to error no-go responses following stress and recognized and correctly withheld no-go responses similar to the control group. Experiencing stress never led participants to ignore the accuracy of the no-go response and perform impulsive and rapid error awareness responses following acute stress.

Errors and Post-error Responses

The number of error trials was 96.27 ± 29.23 in the control group and 111.61 ± 35.20 in the stress group. The number of aware error trials was 62.69 ± 23.41 in the control group and 59.52 ± 28.67 in the stress group. The number of unaware error trials in the control group was 28.65 ± 15.77 and 49.70 ± 32.00 in the stress group. The analysis of the number of error trials in different groups revealed a significant (marginal) effect of Group (P = 0.059), Trial Type (P <0.001), and interaction effect (P = 0.036). Analysis of RTs in aware and unaware error responses showed that no effects were significant (P >0.16). For RTs of the two no-go error responses, a significant main effect of the No-go Type was revealed, F_{(1, 47)} = 12.8, P = 0.001, η_p² = 0.21, as indicated by the RT of the repeat no-go trial (491.56 ± 19.50 ms) being longer than that of color no-go trials (462.10 ± 20.73 ms). The other effects were not significant (P >0.29). Details are presented in Table 2.

The post-error adjustments in accuracy and RTs during the first and second trials after the error response are illustrated in Fig. 4. On the first trial after error response, the Trial Type × Group ANOVA for post-error adjustment in accuracy showed that the main effect of Error Type was significant, F_{(1, 47)} = 5.34, P = 0.025, η_p² = 0.10, indicating that post-error adjustment in accuracy during the post-aware error trials (−1.88% ± 4.80%) was significantly lower than that during the post-unaware error trials (−0.48% ± 2.60%). The main effect of the Group was marginally significant, F_{(1, 47)} = 3.71, P = 0.060, η_p² = 0.07. Moreover, the interaction effect of Trial Type × Group was significant, F_{(1, 47)} = 5.09, P = 0.03, η_p² = 0.10, and further analysis showed that the main effect of Trial Type was only significant in the stress group, F_{(1, 47)} = 9.83, P = 0.003, η_p² = 0.17. In addition, the adjustment in accuracy during the post-aware error trials in the stress group was significantly lower than that in the control groups, F_{(1, 47)} = 5.05, P = 0.029, η_p² = 0.10. Regarding the post-error adjustment in RTs, the main effect of Trial Type was significant, F_{(1, 47)} = 84.22, P <0.001, η_p² = 0.64, with shorter post-error adjustment in RTs during the post-aware error trials (−72.16 ± 65.70 ms) than that during the post-unaware error trials (25.90 ± 33.60 ms). No other effects were significant (P > 0.09). For the second trial after the error response, no significant result was found during post-error adjustment in accuracy (P >0.06). However, during the post-error adjustment in RTs, the Trial Type × Group ANOVA yielded a significant main effect of Trial Type, F_{(1, 47)} = 15.47, P <0.001, η_p² = 0.25, indicating shorter post-error adjustment in RTs during the post-aware error trials (−10.42 ± 30.30 ms) than that during the post-unaware error trials (12.98 ± 23.86 ms). Neither the effect of Group nor the interaction was significant (P >0.13).

Fig. 4.

Two-way ANOVA of post-error adjustment in accuracy and RTs in the control (n = 26) and stress group (n = 23) (mean ± SEM). A The stress group shows lower accuracy during the first trials after the aware error responses than the unaware error responses and the control group. B No significant group difference during the second trial after the error response. *P <0.05, **P <0.01.

ERP Results

ERN and ∆ERN

The Trial Type × Group ANOVA in ERN/CRN following the error and correct responses indicated that no effects were significant (P >0.13). The analysis in ∆ERN showed the same results (P >0.34). The paired-samples t-test for different error awareness responses showed significantly smaller amplitude following aware error responses (0.28 ± 2.11 μV) than after unaware error responses (0.85 ± 2.02 μV, P = 0.03) and correct go responses (0.97 ± 1.42 μV, P <0.01) in the control group. However, this outcome was not found in the stress group (P >0.09) (See Fig. 5).

Fig. 5.

A Response-locked ERN component at the fronto-central region [(FC1 + FCz + C1 + Cz)/4] and topographic maps for correct and error responses in the control (n = 26) and stress group (n = 23). B ERN − CRN difference waves at the fronto-central region and topographic maps for the difference between errors and correct responses in the control and stress groups. The solid line at t = 0 indicates the onset of error response. The orange shadow shows the time window of ERN and CRN (−20 to 30 ms relative to error response). ERN, error-related negativity; CRN, correct-related negativity.

Pe

The repeated-measures ANOVA showed that the main effect of Trial Type was significant, F_{(1.59, 74.65)} = 90.44, P <0.001, η_p² = 0.66. The results indicated the largest positive amplitude following aware error responses (2.47 ± 2.72 μV), the next largest amplitude following unaware error responses (−0.93 ± 3.04 μV), and the smallest amplitude following correct go responses (−2.60 ± 2.49 μV). The waveforms and topographic maps of Pe are displayed in Fig. 6. The Trial Type × Group interaction was also significant, F_{(1.59, 74.65)} = 9.25, P = 0.001, η_p² = 0.16. The interaction was driven by a smaller Pe amplitude (1.35 ± 0.52 μV) following aware errors in the stress group than in the control group (3.47 ± 0.50 μV), F_{(1, 47)} = 8.60, P = 0.005, η_p² = 0.16. The main effect of Group was not significant (P = 0.59).

Fig. 6.

A Response-locked Pe component at the centro-parietal region [(CP1 + CPz + CP2 + Pz + P1 + P2)/6] for correct and error responses in the control (n = 26) and stress group (n = 23). The Pe amplitude following aware error is smaller in the stress group. The solid line at t = 0 indicates the onset of error response. The orange shadow shows the time window of Pe and Pc (250 to 500 ms relative to error response). The same below. B Topographic maps of correct and error responses in the control and stress groups. The stress effect on Pe amplitude is explored by a repeated-measures ANOVA with the factors Error Type × Group. Pe, error positivity. **P <0.01.

ΔPe

The repeated-measures ANOVA for Error Type × Group showed that the main effect of Error Type was significant, F_{(1, 47)} = 53.58, P <0.001, η_p² = 0.53, which revealed that the ΔPe following aware error (5.08 ± 2.88 μV) was significantly larger than that following unaware error (1.67 ± 2.01 μV). Furthermore, the Error Type × Group interaction effect was significant, F_{(1, 47)} = 12.58, P = 0.001, η_p² = 0.21. The simple effect analysis revealed that ΔPe following aware error in the stress group (3.96 ± 2.72 μV) was significantly smaller than that in the control group (6.08 ± 2.68 μV), F_{(1, 47)} = 7.57, P = 0.008, η_p² = 0.14. The ∆Pe following unaware error in the stress group (2.25 ± 2.22 μV) was significantly larger than that in the control group (1.17 ± 1.67 μV), F_{(1, 47)} = 3.56, P = 0.059, η_p² = 0.07. The main effect of Group was not significant (P = 0.30). The waveforms and topographic maps of ΔPe are displayed in Fig. 7.

Fig. 7.

A Pe − Pc difference waves at the centro‐parietal region [(CP1 + CPz + CP2 + Pz + P1 + P2 )/6] in the control (n = 26) and stress group (n = 23). The ∆Pe following unaware error is larger in the stress group. B Topographic maps of the difference between error and correct responses in the control and stress groups. The stress effect on ∆Pe amplitude is explored by a two-way ANOVA with the factors Error Type × Group. Pe, error positivity; Pc, correct positivity. *P <0.05, **P <0.01.

Correlation Analysis

For correlation analysis between the accuracy of error awareness and the error-monitoring component, a significant positive correlation between the accuracy of error awareness and Pe following aware error in the stress group was found, r = 0.33, P = 0.019, but not in the control group (P = 0.33). After correlation analysis of the accuracy of error awareness and accuracy during post-error responses, a significant positive correlation between the accuracy of error awareness and accuracy of the first trial after aware error response in the stress group was revealed, r = 0.44, P = 0.002, as well as in the second trial after aware error response, r = 0.30, P = 0.035, but not in the control group (P >0.19). Furthermore, the Pe following aware error in the stress group shared a significant positive correlation with the accuracy of the first (r = 0.33, P = 0.019) and second trials (r = 0.46, P = 0.001) after the aware error response, but not in the control group (P >0.11). The correlation between behavioral and ERP results with the AUCi of stress responses at each time point of T2–T4 was analyzed, and a false discovery rate (FDR) correction was used when the correlation was significant at two time points. The accuracy of error awareness shared a significantly negative correlation with negative affect at T2 (r = −0.44, P = 0.03) and T3 (r = −0.39, P = 0.06) in the stress group. There was a negative correlation between the accuracy of the first trial after aware error and heart rate at T2 (r = −0.49, P = 0.02) and T3 (r = −0.45, P = 0.03) in the stress group. Otherwise, the amplitude of Pe following aware error was negatively correlated with the cortisol levels at T3 (r = −0.52, P <0.01) and T4 (r = −0.60, P = 0.01) in the control group (Tables S1 and S2). After FDR correction, the correlation between the accuracy of the first trial after aware error and heart rate was significant (P = 0.03), and the correlation between the amplitude of Pe following aware error and cortisol levels was significant (P = 0.006). The small sample size and number of tests in this study might limit the validity of the correlation analysis between the stress indicators and task results.

Discussion

We explored error awareness following acute stress by examining the behavioral and electrophysiological correlates of error processing. The increased salivary cortisol concentration, heart rate, and self-report measures in the stress group indicated successful stress induction. Our results revealed that experiencing stress resulted in lower error awareness rates with faster reactions in individuals. In the stress group, the accuracy of post-aware error trials was lower than that of post-unaware error trials and lower than that found in the control group. All these results are consistent with those reported in our previous study [54]. In the EEG results, the amplitudes of ΔPe and Pe following aware error responses in the stress group were significantly smaller than those in the control group. On the other hand, the amplitude of ΔPe following unaware error responses was significantly greater than that in the control group. Our findings suggest impaired error awareness and post-error performance following acute stress. This experiment revealed the negative impact of impaired error awareness on post-error regulation following acute stress.

The lower conscious level of error responses following acute stress suggests an impaired performance monitoring function. The negative relationship between error monitoring and negative affect under stress was consistent with task disengagement induced by negative emotional states in the previous studies [55, 56]. Attention resources would be relocated to amygdala-driven emotional processing and vigilance following stressful situations, decreasing cognitive resources necessary for goal-directed performance monitoring driven by the higher-order cortex [27] and leading to failure of error recognition. This result of decreased error awareness is also supported by studies that showed impaired performance monitoring under deficient neuropsychological processes for error monitoring in patients with lesions in the frontal (prefrontal) lobe [57, 58]. In addition, it should be mentioned that the faster pressing response to indicate an aware error is caused by increased excitability in the motor cortex following stress [28, 59]. Faster Go responses in the stress group and no significant group difference in the P_{false awareness} indicated that stress-induced motor preparation leads to a faster pressing response.

We found that the ERN was regulated by conscious error perception only in the control group. Based on previous findings, the ERN component is sensitive to early automatic detection of errors and acts as a feedforward input signal for systems more in charge of error awareness [60]. The conscious perception of errors seems already impaired during the initial detection phase following stress. Moreover, there was a typical phase of the Pe following the aware error responses while being absent following the unaware error responses in the control group. Consistent with previous findings [22, 39], the Pe component indicated the evidence accumulating for error signals to generate error awareness. Conscious error perception is generated by the integration of top-down attention to salient signals and bottom-up representation of information about error responses from various sources. Evidence accumulating for error signals generated conscious error perception while lacking evidence accumulating induced incognizant perception presented as no Pe amplitude. Notably, the amplitude of Pe following the aware error responses in the stress group was significantly smaller than that in the control group, while the amplitude of Pe following the unaware error responses in the stress group was greater than that in the control group. This result indicated that the error identification was based on less evidence accumulation in the stress group, and error evaluation also occurred under unidentified errors. As previously stated, the neural basis for evidence accumulation, such as the pMFC, insula, and PFC, is susceptible to stress and stress responses. In particular, the PFC is the core brain area for task representation, and stress-induced dysfunction would lead to insufficient evidence accumulation. Furthermore, the global workspace theory proposes that the changes in Pe amplitude reflect the content of the global workspace of error awareness [61], including factors such as task context and motor reaction. A study by Di Gregorio, Steinhauser, and Maier (2016) also found that the amplitude of Pe under partial error awareness was smaller than that under complete error awareness [19]. Other research also revealed that confidence in the error response mediated the Pe amplitude [62], which manifested adequate error assessment as a larger Pe amplitude. Our experiment suggested that stress-induced inadequate evidence integration in the global workspace following acute stress generated a more ambiguous and uncertain error identification.

This experiment revealed faster responses after identifying erroneous responses. It is a common phenomenon of speedup after error no-go responses in Go/No-Go tasks [30, 63]. Since pressing the Go button is a prepotent response, the participant automatically processes the prepotent response after the error response. Moreover, decreased accuracy in post-error trials in both groups was mediated by a relatively short response stimulus interval (700 ms). The maladaptive accounts of post-error adjustment suggest that error monitoring temporarily interferes with the processing of subsequent trials by occupying its cognitive resources [11, 64, 65]. Most experiments that showed post-error accuracy improvement had inter-trial intervals of >900 ms [66, 67]. More importantly, individuals in the stress group exhibited impaired post-aware error regulation, and their performance was worse than that of the control group. Acute stress seemed to generate interference with the generation and performance of post-error regulation. On the one hand, the positive correlation between error awareness rates and accuracy of the trials after aware error response in the stress group indicated that performance monitoring and post-error adjustment shared similar neuropsychological processes following stress. Dysfunctional higher-order functions following acute stress [8, 68] weaken error monitoring and impair post-error adjustment. The top-down regulation of attention, which suffers from the stress effect [69], might make it harder for individuals to appropriately improve task performance following error responses, as factors such as locating attention to task-related information might be impacted. On the other hand, inadequate evaluation and discrimination of error responses following acute stress might impair factors of the initiation of processing of post-error trials, such as perceptual information processing. An ambiguous and resource-intensive assessment of ongoing performance might prolong the bottleneck in processing post-error stimuli, resulting in inefficient processing in subsequent trials.

Previous studies have shown a negative impact of stress on error processing, especially error monitoring. Here, we further examined the impact of error awareness on error processing and revealed the full-stage feature of error processing under stress. Based on the behavioral and EEG results, we demonstrated impaired error awareness following acute stress, which induces inappropriate post-error behavior regulation. The dual competition theory proposes that emotional processing and executive control compete for limited cognitive resources. When intense emotional states occupy cognitive resources, executive control processing is negatively affected [70]. More specifically, the biphasic‐reciprocal model also presents a reallocation of cognitive resources to the salience network, promoting vigilance and fear at the cost of the executive control network under stress [26]. In this way, a state dominated by vigilance following acute stress impairs error processing through the poor function of a higher-order performance monitoring system and further top-down regulation. On the one hand, stress-induced vigilance and fear occupy more cognitive processing resources, which causes inadequate error identification in the response monitoring stage. On the other hand, impaired error monitoring interfered with post-error response adjustment generation and initiation, resulting in inappropriate post-error adjustment. Notably, an error response is a typical negative signal for individuals. Emotional responses help generate adaptive response preparation for aversive signals (errors) and play an essential role in impelling the regulation of post-error responses [71]. Intense negative experiences under stress do not promote the generation of error awareness but further interfere with subsequent behavior adjustment. Our research revealed the impairment in a series of stages of error processing following stress. Impaired error monitoring under stress is crucial in post-error maladaptive regulation. Impaired error awareness following stress provides insight into the behavioral adaptation of stress and post-traumatic stress disorder patients and offers options for further therapeutic approaches.

Several limitations of this research must be considered. First, to avoid the effects of gender and the menstrual cycle on stress responses, all participants in this study were male. Some evidence has shown differences between male and female participants in cognitive processing after experiencing stress [46, 72]. Therefore, future studies need to consider the influence of gender on the stress effect. Second, the sensitivity of the ERN component for error awareness varies in different paradigms [73]. Here, we revealed a weak ERN effect via a typical Go/No-Go error awareness task, and the specific stress effects in other paradigms still need to be explored. Finally, the correlation between the Pe components and cortisol in the control group is consistent with the studies that found error monitoring related to baseline cortisol [74, 75]. However, these studies did not report the results of Pe components, and further research is needed to explore what factors may cause variations in the outcomes.

In conclusion, we showed that acute stress impairs different stages of error processing. Acute stress had a negative effect on error monitoring, manifested in a low level of conscious error recognition, small ΔPe amplitudes following aware error responses, and increased ΔPe amplitudes following unaware error responses. At the same time, inadequate error identification led to worse post-error performance following stress. We revealed the overall perspective of the stress effect on different stages of error processing. These results suggested that impaired error awareness following stress drives maladaptive behavioral adjustment.

Supplementary Information

Below is the link to the electronic supplementary material.

Data Availability

Data supporting the findings of this study are available from the corresponding author upon reasonable request.

Conflict of interest

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