Mapping emotion-modulated inhibitory control in borderline personality features: a dimensional approach using the emotional Go/No-Go task with EEG

Abstract

Background

To explore how emotional-modulated inhibitory control, as assessed by the emotional Go/No-Go task and EEG, correlates with the multidimensional profile of borderline personality disorder (BPD).

Methods

Eighty-two participants completed the Personality Assessment Inventory-Borderline Features (PAI-BOR), Patient Health Questionnaire-9 (PHQ-9), and Generalized Anxiety Disorder-7 (GAD-7) scale. EEG was recorded during an emotional Go/No-Go task and event-related potential (ERP) components were analyzed for correlations with BPD features. Participants were then divided into groups with or without BPD features, and comparisons of ERP components were then made within groups.

Results

ERP analysis revealed BPD features correlations with brain activity. Affective instability was positively correlated with left frontal N2 during positive-neutral No-Go trials. Self-harm was associated with left frontal late positive potential(LPP) during negative-neutral Go trials. Left frontal and central LPP slightly related to identity problems. 32 participants grouped into participants with BPD features(the BPF group) showed more depression, anxiety, emotional instability, identity issues, self-harm, and interpersonal problems. The BPF group had impaired performance on No-Go trials, smaller N2 amplitudes at left frontal channels during negative emotional cues, and delayed N2 peak latency. A group-trial type interaction was observed, with higher LPP amplitudes in Go trials for the control group but not in the BPF group, indicating distinct emotion processing between the BPF and control groups.

Conclusion

The findings of this study propose that there are neural associations among affective instability, self-harm, and identity problems in BPD, which are consistent with the neural foundation of emotion-modulated inhibitory control observed in the emotional Go/No-Go task. The lack of association between negative sociality and ERP components highlights the complex nature of social information processing in BPD. Moreover, significant emotional, cognitive, and neural differences are also observed between the BPF and control groups. These results enhance understanding of how emotional valence modulates inhibitory control processes in individuals with varying levels of BPF.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40479-025-00325-z.

Background

Borderline personality disorder (BPD) is a complex mental health condition characterized by a pervasive pattern of emotional instability, impulsivity, and interpersonal difficulties [1]. The prevalence of BPD is approximately 0.7%-2.7%, with typical onset during early adulthood [2]. BPD is associated with a high rate of suicidal behavior, with an attempted suicide rate of up to 75% and a completed suicide rate of approximately 8%-10%, which is 50 times higher than the general population [3]. In a 24-year prospective study, 5.9% of individuals with BPD subjects died by suicide [4]. Research indicates that individuals with BPD have a shortened life expectancy by 6–7 years compared to the general population [5, 6].

BPD is usually diagnosed using structured clinical interviews, and the Diagnostic and Statistical Manual of Mental Disorders lists nine diagnostic criteria for the disorder. The typical features of BPD include emotional dysregulation, identity disturbances, impulsivity, and unstable interpersonal relationships [7]. Emotional dysregulation, the core feature of BPD, is characterized by intense and rapidly fluctuating emotional states, difficulties in modulating emotional responses, and a limited capacity to tolerate emotional distress [8]. BPD is also marked by identity disturbances, reflected in an unstable and fragmented sense of self, with significant shifts in self-perception, values, goals, and relationships [9, 10]. Additionally, impulsivity is a prominent feature of BPD, manifesting through risky behaviors such as substance abuse, unprotected sexual activity, binge eating, reckless driving, and non-suicidal self-injury [11, 12]. Unstable interpersonal relationships are also a hallmark of BPD, with a pattern of intense and rapidly shifting relationships, fear of abandonment, and difficulties maintaining healthy boundaries [13]. Previous studies have developed a lot of self-report questionnaires like the Borderline Personality Questionnaire, Borderline Symptom List-Short Form, and the Difficulties in Emotion Regulation Scale and Personality Assessment Inventory-Borderline Features (PAI-BOR) to evaluate different aspects of the disorder [7, 14, 15]. However, questionnaires can be subject to biases and may not fully capture the underlying cognitive and emotional processes associated with BPD. Therefore, objective measures, such as cognitive tasks and neuroimaging techniques, are needed to provide complementary information and gain a more comprehensive understanding of the neurobiological underpinnings of BPD. Moreover, diagnostic boundaries can be artificial; recent literature emphasizes that psychopathological features, including those features of BPD, exist on a continuum within the general population [16, 17]. Examining continuous features profiles rather than employing binary diagnostic categories facilitates a more nuanced understanding of symptom severity and heterogeneity. Hence, this dimensional approach helps identify specific neural correlations underlying distinct BPD feature dimensions, which may be masked in categorical comparisons.

To comprehensively understand the neurocognitive mechanisms underlying continuous features in BPD, researchers have employed experimental paradigms such as the emotional Go/No-Go task, which directly assesses emotional inhibitory control—a critical feature of emotional dysregulation for BPD. The emotional Go/No-Go task is a behavioral measure designed to assess an individual’s ability to inhibit a prepotent response in the context of emotionally salient stimuli. This allows for the examination of the neural and behavioral correlates of response inhibition in emotionally charged contexts, capturing how emotion processing and inhibitory control interact to address emotional dysregulation in illicit and excited states [18]. Individuals with BPD have demonstrated impaired performance on this task, with increased false alarms (failures to inhibit a response) to negative emotional stimuli compared to the control group, suggesting difficulties in emotional inhibitory control [19]. However, the relationship between the emotional Go/No-Go task performance and specific BPD feature dimensions remains unclear. Specifically, event-related potential (ERP) measures obtained during the emotional Go/No-Go task can provide valuable insights into the underlying neural processes associated with impulsivity and emotion-modulated inhibitory control deficits in BPD. N2 typically occurs between 200 and 350 ms following stimulus presentation, reflecting conflict monitoring and appearing in No-Go trials. Shorter N2 latencies and lower global connectivity were observed in the BPD patients regardless of the emotional valence of the background images compared to controls [20]. However, there has been inconsistently results in amplitude between BPD patients and controls in the Go/No-Go task in previous research [21–23]. These inconsistencies may arise from variations in personality measures or data analysis methods. Another component, P3, typically occurring between 300 and 500 ms, reflects cognitive evaluation and resource allocation. On No-Go trials, P3 reflects inhibition processes and the evaluation of inhibitory performance [24]. Research indicates that while both BPD patients and healthy controls showed larger frontocentral P3 amplitudes for No-Go trials compared to Go trials, the magnitude of this effect was demonstrably greater in the control group than in the BPD group [24]. This attenuated P3 response in BPD patients during inhibitory control tasks suggests a diminished capacity for cognitive evaluation and resource allocation during response inhibition, which aligns with the emotion-modulated inhibitory control observed in individuals with BPD. Moreover, late positive potential (LPP), as a component reflecting sustained attention and emotional processing, was more positive in the BPD group than the control group, and the negative stimuli could elicit larger LPP amplitude than positive stimuli, which meant that BPD patients paid more attention to negative stimuli [25, 26]. Hence, these ERP findings complement the behavioral impairments observed in the emotional Go/No-Go task, offering a deeper understanding of the neurocognitive mechanisms driving the emotional and behavioral dysregulation characteristic of BPD [26]. For instance, negative emotional stimuli may lead to increased neural reactivity in individuals with BPD, as shown by heightened LPP amplitudes. This heightened emotional processing of aversive cues may contribute to features like affective instability and self-harm. On the other hand, positive emotional stimuli may challenge inhibitory control in BPD, potentially leading to difficulties in suppressing responses to reacting cues. This could be reflected in alterations in the N2 component, which is involved in conflict monitoring. However, further research is needed to clarify the specific relationships between the emotional Go/No-Go task performance, underlying neural processes, and the multidimensional features of BPD.

The goal of this study is to investigate the associations between emotion-modulated inhibitory control measured behaviorally and through ERP components—and the dimensional profile of BPD features. By adopting a dimensional perspective across a non-clinical sample, we seek to clarify how specific feature dimensions relate to underlying neural mechanisms. Examining these relationships can enhance our understanding of the mechanisms driving the emotional and behavioral difficulties observed in individuals with BPD features. To further support the dimensional analysis findings of BPD, this study will explore the behavioral and electrophysiological links of emotional inhibitory control in individuals with borderline personality features (BPF). We predict that ERP components such as N2, P3 and LPP will show valence-specific alterations in the BPF group, suggesting unique neural processes underlying emotion-modulated inhibitory control in various emotional contexts.

Methods

Participants

From April 2024 to October 2024, recruitment posters were distributed to prospective participants at three universities (refer to Supplementary Fig. 2 for a comprehensive depiction of the participant enrollment and grouping procedures). The posters included the link for completing the PAI-BOR scale and also summarized the purpose and main details of the study. We collected 737 questionnaires using a convenience sampling method. All these students were then invited to participate in the experimental phase, with detailed explanations of the procedures and requirements provided. The inclusion criteria were university students aged between 18 and 25 years who gave informed consent and participated voluntarily. Exclusion criteria included: reading difficulties or sensory impairments; neurological disorders, head injuries or severe physical illnesses; uncorrected visual acuity below 1.0; any physical discomfort in the week prior to the experiment; current or past diagnosis of impulse control disorders (e.g., intermittent explosive disorder), or Attention-Deficit/Hyperactivity Disorder; current diagnosis or recent history (within the past year) of any other Axis I psychiatric disorder (assessed via structured clinical interview); medications in the past 6 months; and prior participation in similar experiments. Moreover, the handedness of participants was not specified, and the experiment instructions did not require the use of the right hand exclusively for key pressing. Participants were allowed to choose whichever hand position they found most comfortable to complete the task. In order to boost participation, participants were given compensation after finishing the experiment. Subsequently, we screened the respondents based on specific criteria, resulting in 89 students volunteering to participate in the task. Following data cleaning and quality control, the final analysis included data from 82 participants who met all criteria. Post hoc power analyses confirmed that the achieved sample size of 82 participants yielded sufficient statistical power (power > 0.90) to detect moderate-to-large effects in both the factorial ANOVA and regression models, thereby ensuring the robustness of the primary findings.

Measures

Demographic and clinical measures

A demographic questionnaire was developed based on a review of the relevant literature. It includes items assessing participant age, gender, and their schools. To assess clinical symptoms, the study utilized two well-established self-report measures. The Patient Health Questionnaire-9 (PHQ-9) is a commonly used self-report measure for evaluating depressive symptoms severity and aiding in depression diagnosis [27]. It assesses the frequency and severity of nine depressive symptoms over the past two weeks. In our research, the Cronbach’s alpha coefficient for the PHQ-9 was 0.89. The Generalized Anxiety Disorder-7 scale (GAD-7) is utilized for screening and evaluating generalized anxiety symptoms [28]. It evaluates the frequency of seven anxiety-related symptoms experienced in the past two weeks. In our research, the Cronbach’s alpha coefficient for GAD-7 was 0.93.

Borderline personality assessment

The Personality Assessment Inventory - Borderline Features (PAI-BOR) scale is a subscale of the broader Personality Assessment Inventory [29]. The scale is designed to assess personality features and psychopathology, with a specific focus on features of BPD. The PAI-BOR scale comprises four subscales, providing a more nuanced evaluation of BPD features:

• Affective Instability: Measures emotional lability, intense emotional experiences, and difficulties with emotion regulation.

• Negative Sociality: Evaluates difficulties in interpersonal relationships, including mistrust, hostility, and a tendency to perceive others negatively.

• Self-Harm: Assesses the presence and frequency of self-harming behaviors, including suicidal thoughts and actions, as well as non-suicidal self-injury.

• Identity Problems: Assesses confusion or uncertainty about self-image, values, and long-term goals.

The PAI-BOR scale comprises 24 items with a 4-point Likert-type response format ranging from 0 to 3. A score of 38 or higher on the PAI-BOR scale suggests the presence of BPD, with higher scores indicating a greater likelihood and serving as a criterion for BPF [30]. PAI-BOR scale is widely used measure of BPD features in undergraduate samples and associated with interview-based diagnoses of BPD [31, 32]. Internal consistency was excellent for the overall scale (α = 0.92) [33]. The PAI-BOR demonstrated good internal consistency in our sample, with alpha coefficients ranging from 0.75 to 0.85 for the subscales.

Emotional Go/No-Go task

The emotional Go/No-Go task (Fig. 1) was adapted from the paradigm developed by Zutphen et al. [34]. The stimuli were selected from the International Affective Picture System and consisted of 60 pictures, representing three emotional categories (negative, neutral, and positive) with 20 pictures in each category. The neutral stimuli were selected to match the mean luminance and complexity of the pictures. Given the heightened sensitivity of individuals with the BPD features to interpersonally relevant content, only emotional pictures with interpersonal relevance were included. The emotional valence and arousal properties were thoroughly evaluated in Supplementary Table 1. The pictures were presented in two different formats: Go stimuli had a blue frame, and No-Go stimuli had a yellow frame. Participants were instructed to quickly and accurately press the “J” key on the keyboard when they saw a Go stimulus, and to refrain from responding when they saw a No-Go stimulus. Participants were not required to use a specific hand when pressing the “J” key, allowing them to use whichever hand they found most comfortable. The experiment consisted of 360 trials (20 pictures × 3 emotions × 2 frames × 3 runs), with 70% Go stimuli and 30% No-Go stimuli. The two presentation formats of the three emotional picture types were combined to form 6 blocks, each containing 20 trials of the two emotional picture types (with a 7:3 ratio of Go to No-Go stimuli). The blocks were pseudo-randomized in terms of stimulus presentation order within each block to reduce predictability. Participants were required to take at least a one-minute break between blocks to prevent fatigue. Each run consisted of 6 blocks, and the entire experiment had 3 runs. Before the official experiment, participants completed 12 practice trials to familiarize themselves with the task instructions. Each trial consisted of a fixation point presented for 1200–1500 ms, followed by a 1000 ms stimulus presentation (with the stimuli appearing in a pseudo-random order), and finally a 1200–1500 ms blank screen. The experiment was programmed using E-Prime 2.0, which recorded the participants’ reaction times and accuracy. The experiment was conducted in a quiet and comfortable electroencephalography laboratory, and participants were instructed to remain alert and attentive throughout the task.

Fig. 1.

The emotional Go/No-Go task design. Note: Panel A shows a block in which negative pictures were combined with Go-trials and positive pictures with No-Go-trials. Each trial consisted of a fixation point presented for 1200–1500 ms, followed by a 1000 ms stimulus presentation (with the stimuli appearing in a pseudo-random order), and finally a 1200–1500 ms blank screen. Panel B depicts the order of the blocks and set Go/No-Go-combinations

EEG acquisition and preprocessing

The electroencephalography data in this study were collected using a NeuroScan EEG system, with 64-channel electrode caps based on the extended international 10–20 system to record the EEG signals. During the data acquisition, the system’s built-in “REF” electrode was used as the reference, which was later converted to an average reference during offline analysis. The recording commenced when the impedance between the reference electrode and the scalp was below 5 kΩ, and the impedance between the remaining electrodes and the scalp was below 10 kΩ. The sampling rate was set to 1000 Hz. Participants were instructed to remain alert and attentive throughout the data collection process and were asked to minimize eye movements and head movements as much as possible. Preprocessing was performed using the EEGLAB toolbox and the details of this preprocessing approach have been described in detail in Debnath et al. [35]. First, the acquired EEG data were imported into the EEGLAB toolbox, and the sampling rate was set to 500 Hz. The EEG data underwent bandpass filtering using a zero-phase Butterworth filter (0.5–45 Hz). Following this, independent component analysis was utilized to detect and eliminate components associated with eye movements, muscle artifacts, and other sources of noise. Subsequently, a semi-automatic rejection method was applied with a threshold of ± 100 uV. Finally, the EEG data were segmented into epochs ranging from − 200 ms to 1000 ms relative to the onset of stimulus presentation. Baseline correction was applied using the − 200 to 0 ms pre-stimulus interval. Epochs with excessive noise or artifacts were automatically excluded using the HAPPE 2.0 plugin. Further information on EEG preprocessing outcomes regarding trial rejection can be located in the supplementary excel file(preprocess for each condition.xlsx). Epochs were extracted for the following conditions: Go trials with negative, neutral, and positive cues, as well as No-Go trials with negative, neutral, and positive cues. For each participant, the researchers computed the mean value across all available epochs to create an average epoch for each condition. This process was repeated for each electrode channel. The difference waves were calculated by subtracting the neutral condition from the negative and positive conditions, separately for the Go and No-Go trials, in order to contrast the emotional conditions. Averaged epochs were grouped together into scalp-based regions of interest (ROIs). This was accomplished by calculating the mean across clusters of electrode channels: namely, frontopolar (FPz, FP1, FP2, AF3, AF4) involved in abstract reasoning and goal-directed behavior; left frontolateral (AF7, F5, F7) and right frontolateral (AF8, F6, F8) implicated in executive functions; frontocentral (FCz, Cz, FC1, FC2, C1, C2) assessing motor control, response monitoring, and error processing; occipitalmedial (Oz, POz, O1, O2, PO3, PO4) measuring visual processing, attention, and early sensory activity; left posteriorlateral (CP5, CP7, CP9, P5, P7, P9) and right posteriorlateral (CP6, CP8, CP10, P6, P8, P10) investigating spatial processing, attention, and sensory integration. Usually, N2 component is observed in the frontolateral brain regions, typically peaking around 200–300 milliseconds after stimulus onset [36]. Simultaneously, the P3 component is predominantly detected in centroparietal areas, with a peak latency around 300–500 milliseconds following the stimulus presentation [37, 38]. The LPP, which overlaps temporally with the P3 but typically lasts longer, is also centered over centroparietal to parietal regions, often persisting from 300 to 1000 milliseconds, particularly in response to emotionally salient stimuli [38, 39].

Statistical analyses

These analyses have examined behavioral data and EEG data separately:

For the behavioral data, reaction times and accuracy rates will be calculated for Go and No-Go trials, with separate analyses for the three emotional conditions (positive, negative, and neutral, marked by“-”, “|” and “+”). A correlation analysis was conducted to assess the relationship between behavioral performance and PAI-BOR scores. To provide a more rigorous analysis, the dataset was reanalyzed using ANOVAs, with the BPF (participants reported high BPD features and scored equal to or above 38 of PAI-BOR) and control (participants reported low BPD features and scored below 38 of PAI-BOR) group category as a between-subjects factor and emotional conditions as a within-subjects factor for both reaction times and Go/No-Go accuracy. The effects and interactions resulting from these analyses are reported in Supplementary Tables 2–4. Post hoc t-tests were conducted with Bonferroni correction for multiple comparisons to further analyze the effects of emotional condition and BPD diagnosis on task performance.

For the epoched EEG data, correlational analyses and regression models will be employed to examine the relationships among BDP features, response inhibition under emotional processing as measured by the emotional Go/No-Go task, and relevant EEG metrics. To investigate the relationship between ERP activity and BPD features, a regression model was calculated at each time point (1000 milliseconds post-stimulus onset). This analysis accounted for potential confounding factors, such as sex, and age, by including them as covariates. To address potential multicollinearity between predictors, particularly given the inclusion of both linear and quadratic ERP terms, we assessed variance inflation factors for each predictor within the regression models. VIF values were consistently below 3, indicating that multicollinearity was not a significant concern. Both the predictor and outcome variables were standardized to z-scores. The model calculated at each time point could be as follows:

Φ_{BPD, feature} represents the score for a specific feature or a composite score of BPD that is modeled separately for each subscale or overall BPD features. β_{negative−neutral, ERP} and β_{positive−neutral, ERP} represent the regression coefficients that quantify the difference in event-related potential activity between negative and positive emotional cues, respectively, compared to neutral cues. β_{negative−neutral, ERP}² and β_{positive−neutral, ERP}² represent the regression coefficients for the curvilinear (quadratic) relationship between ERP activity and BPD features. β_age, and β_sex are regression coefficients for the control variables.

The statistical significance of the regression model terms was determined using two approaches: a maximum statistic approach with permutation models and an evaluation of the number of contiguous time points surpassing a significance threshold. That is, for every ROI, 1000 null models were also calculated by randomly shuffling the EEG predictors across participants before recalculating the regression model. This created a set of 7000 null models at each time point (1,000 null models x 7 ROIs). Pooling all the beta values derived from the permutation models, the 95th percentile of the distribution was identified. This method was applied to [1] identify significant beta values within each model and [2] identify ROIs in which the R-squared (R2) value was significantly greater than chance [40]. In addition to the permutation-based correction, an evaluation was performed of the number of contiguous significant time points exceeding a predefined significance threshold to assess the robustness of the effects. The second approach involved determining statistical significance based on the number of consecutive time points where a model term exceeded the maximum statistical significance threshold. If the maximum statistic approach identified a significant finding, the number of contiguous significant time points were counted. This total was then described in terms of the probability of it occurring in all 7000 null models, which was referred to as the ‘contiguous p’ value. We selected ROI based on prior literature to reinforce the reliability of our results [40]. Contiguous significance clusters were defined with a threshold of p < 0.05 and a minimum cluster size of 50 points to correct for multiple comparisons. The cluster size was determined using a permutation-based approach with 1000 permutations to control the family-wise error rate at p < 0.05. These analyses were exploratory, not confirmatory, with no strong prior hypotheses. This transparency frames the findings as preliminary or hypothesis-generating, rather than conclusive. This dual examination can lead to a broader cluster window where maximal statistical effects are observed (reflecting the cluster-based threshold used for effects with significant beta values) and a narrower time window (often used for plotting regression R² values). Tests were calculated using MATLAB. Data and processing scripts are available online (https://osf.io/uw2s4/?view_only=7286b6067dd2454b8c350dc2f7252947).

Results

The connection among individual characteristics, behavioral indexes, and BPD features

The study population consisted of 82 participants, with 46.3% being male (n = 38) and 53.7% female (n = 44). The average age of the participants was 21.71 years, with a standard deviation of 2.36 years, suggesting a sample of young adults mostly between 19 and 24 years old, as all the participants were college students. Gender had a slight positive correlation with affective instability (r = 0.227, p = 0.040) according to Fig. 2, whereas age did not exhibit significant correlation with other variables. The correlation matrix in Fig. 2 revealed strong evidence of interconnected relationships among individual characteristics, behavioral indices, and BPD features, highlighting the interwoven nature of these variables. Affective instability, identity problems, negative sociality, and self-harm all displayed strong correlations with PHQ (depressive symptoms), GAD (anxiety), and PAI-BOR (general psychopathology). The correlation coefficients ranged from moderate to very high (with r ranging from 0.6 to 0.9), all highly significant (p < 0.001). Positive-Go (GO+) ACC was negatively correlated with PAI-BOR, affective instability, and negative sociality (with r ranging from − 0.2 to -0.3, p < 0.05), indicating that poorer behavioral performances were associated with increased psychopathology. Similarly, positive-No-Go (No-GO+) ACC and neutral-No-Go (No-GO|) ACC were negatively correlated with all BPD features, indicating potential impairments in inhibition or decision-making among individuals with higher levels of psychopathology. High self-reported psychopathology was linked to decreased accuracy in Go/No-Go tasks, particularly those involving inhibition and control.

Fig. 2.

Demographic and psychological characteristics of participants with Borderline Personality Disorder (BPD) Compared to control group. Note: N: Number of participants; PHQ: Patient Health Questionnaire, measuring depression symptoms; GAD: Generalized Anxiety Disorder scale, measuring anxiety symptoms; PAI-BOR: the Personality Assessment Inventory - Borderline Features; ACC: Accuracy (proportion correct responses); RT: Reaction Time (in milliseconds); Values are presented as means ± standard deviations for continuous variables and counts (percentages) for categorical variables. Stimuli for Go/No-Go trials were indicated as “–” for negative, “|” for neutral, and “+” for positive. *: p < 0.05, **: p < 0.01, ***: p < 0.001

ERP associations with BPD features

Figure 3 illustrates the grand average ERP waveforms recorded at seven ROIs under various stimulus conditions. Distinct ERP components—N2, P3, and LPP—were clearly visible at specific scalp locations. The N2 component typically appeared between 150 and 300 ms post-stimulus in posterior regions (medial occipital and posterior lateral right/left), while the P3 component followed at approximately 300–400 ms in these same posterior channels. In contrast, at frontal sites (frontolateral right/left and frontocentral), the N2 occurred slightly later, between 200 and 400 ms, and was succeeded by the LPP component observed between 400 and 800 ms. The timing and spatial distributions indicate varying neural processing stages in response to the stimuli. Regression analysis was then used to explore the connection between emotional ERP responses from different waves in emotional and neutral stimuli and BPD features, while controlling for age and sex. Results from the maximum statistic approach showed significant associations between ERP activity during the emotional Go/No-Go task and specific BPD features.

Fig. 3.

Grand averages at 7 ROIs across various conditions and Go/No-Go trials

Affective instability

The analysis using the maximum statistic approach revealed that the critical R-squared threshold for the entire epoch was approximately 0.256 ± 0.001. Furthermore, the probability of observing 50 or more consecutive samples with a significant R-squared value was less than one in 7,000 models indicating the model might be capturing something real. Based on these criteria, the study found that ERP activity in two specific scalp regions, the left frontal lateral and occipital medial areas, were significantly associated with affective instability (see Fig. 4A). The findings indicated that the affective instability was positively correlated with ERP responses in the left frontal lateral region during the positive-neutral-No-Go (No-GO|+) trials, specifically within the 68 ~ 232 ms time frame (β_{range =} 0.382 ~ 0.475; β_threshold (M ± SD) = 0.439 ± 0.025, contiguous p < 0.001). The significant linear regression effect existed from 152 to 240 ms (R² _{range =} 0.255 ~ 0.286), a period corresponding to the N2 component as illustrated in the top panel of Fig. 4B. Additionally, affective instability was inversely associated with ERP responses in the occipital medial region during the negative-neutral-Go (GO|–) trials, occurring within the 224 ~ 362 ms time frame (β_range = -0.369~-0.291; β_threshold (M ± SD)= -0.329 ± 0.024, contiguous p < 0.001). The significant linear effects existed from 248 to 272 ms (R²_{range =} 0.257 ~ 0.265), a period corresponding to the N2 component as shown in the bottom panel of Fig. 4B. The other associations between ERP responses in the left frontal lateral and the occipital medial region during the other cue were not statistically significant (refer to Supplementary Fig. 2 and Fig. 3).

Fig. 4.

The average contrasting waves in response to Go and No-Go cues under different emotional conditions (negative vs. neutral and positive vs. neutral) and their associations with affective instability. Note: In panel A, each panel represents a scalp ROI, with the lighter region surrounding the ERP amplitudes indicating the standard deviation. The colored vertical bars indicate periods when the amplitudes of waves were significantly associated with features, based on R² values exceeding the top 97.5 percentile of all null model R² values. The R² value at each time point is shown at the bottom of each ROI panel, with the color corresponding to the scalp ROI. The dashed blue line represents the critical R² threshold based on the distribution. In panel B, significant periods of ERP activity correlating with features are indicated by colored vertical bars in each panel, determined by regression analysis. Colored horizontal lines indicate linear association with yellow solid lines, curvilinear association with yellow dashed lines, upper and lower threshold for the association with dark solid and dashed lines separately, and the statistical p values for these associations with blue solid and dashed lines separately. The dot and short-dash blue lines show critical thresholds for statistical significance. Statistical significance was assessed using a maximum statistical approach, with benchmarks established from 7,000 null models to ensure robust findings. Thess p-values was log transformed so that higher values trended towards statistical significance: where p = 0.05, -log10 (p) = 1.30; where p = 0.01, -log10 (p) = 2; where p = 0.001, -log10 (p) = 3

Negative sociality

The critical R² threshold across the entire epoch was approximately 0.219 ± 0.001. Based on the probability of observing consecutive samples, ERP activity showed no significant association with negative sociality in any of the ROIs (see Supplementary Fig. 4).

Self-harm

The maximum statistical analysis revealed a critical R-squared threshold of approximately 0.226 ± 0.001 across the entire epoch and a stretch of consecutive data points appear to show a significant R-squared value just by random chance. The study found that only the left frontal lateral ROI was closely linked to self-harm in the participants (see Fig. 5A). The findings indicated that the self-harm was positively correlated with ERP responses in the left frontal lateral region during the Go|– trials, specifically within the 660 ~ 786 ms time frame (β_range = 0.289 ~ 0.382; β_threshold (M ± SD) = 0.339 ± 0.030, contiguous p < 0.001). The significant linear regression effect existed from 674 to 704 ms (R²_range = 0.225 ~ 0.232), a time window corresponding to the late positive potential (LPP) component, as depicted in Fig. 5B. The other associations between ERP responses in the left frontal lateral region during the other cue were not statistically significant (refer to Supplementary Fig. 5).

Fig. 5.

The average contrasting waves in response to Go and No-Go cues under different emotional conditions (negative vs. neutral and positive vs. neutral) and their associations with self-harm. Note: The designation of each component is akin to that of Fig. 3

Identity problems

The highest statistical analysis showed that the critical R-squared threshold for the entire epoch was around 0.215 ± 0.001. Additionally, the likelihood of seeing 50 or more consecutive samples with a significant R-squared value was less than one in 7,000 models. Based on these criteria, the study found that ERP activity in two specific scalp regions, the left frontal lateral and frontal central areas, were significantly correlated with identity problems as depicted in Fig. 6A. The findings indicated that the severity of identity problems was positively correlated with ERP responses in the left frontal lateral region during the Go|– trials, specifically within the 664 ~ 774 ms time frame (β_range = 0.320 ~ 0.412; β_threshold (M ± SD) = 0.383 ± 0.026, contiguous p < 0.001). However, the expected significant linear regression effect was not observed within the 664 to 774 ms time window, where the R-squared values ranged from 0.195 to 0.197 (see Fig. 6B). Additionally, identity problems were positively associated with ERP responses in the frontal central region during the No-Go|+ trials, occurring within the 630 ~ 734 ms time frame (β_range = 0.296 ~ 0.350; β_threshold (M ± SD) = 0.330 ± 0.015, contiguous p < 0.001, see the top panel of Fig. 6B). However, the significant linear effects did not exist and (R²_range = 0.177 ~ 0.188, see the bottom panel of Fig. 6B). The other associations between ERP responses in the left frontal lateral and the occipital medial region during the other cue were not statistically significant (refer to Supplementary Figs. 6 and 7).

Fig. 6.

The average contrasting waves in response to Go and No-Go cues under different emotional conditions (negative vs. neutral and positive vs. neutral) and their associations with self-identity. Note: The designation of each component is akin to that of Fig. 3

Differences between the BPF and control groups in self-reported and behavioral measures

Based on the PAI-BOR criterion, 32 out of 82 individuals exhibit high BPD features and are classified into the BPF group. Statistically significant findings showed that the BPF group had higher mean scores on the PHQ and GAD measures (ps < 0.001, see Supplementary Table 2 for details), indicating greater levels of depression and anxiety compared to the control group. As depicted in Fig. 7, the extent to the BPF group consistently show elevated presence across multiple feature domains. In contrast, the control group are more sparsely represented across the subscales, especially in domains such as Self-Harm and Affective Instability. This observation illustrates the clustering of features within the BPF group and underscores the heterogeneity in psychopathology.

Fig. 7.

Comparison of features between BPF and control groups. Note: The flows (alluvia) trace the pattern of category membership across subscales. BPD-Borderline Personality Disorder. The vertical axis reflects the value associated with each group, which represents either the weighted score for scales. Wider strata and flows indicate a greater number or higher scores of individuals within that category and subscale

Results of the repeated-measures ANOVAs for reaction time of Go-trials did not show significant main effects(see Supplementary Table 3). However, for accuracy rates, there were significant main effects of group (F = 8.78, p = 0.004) and Go/No-Go type (F = 21.40, p < 0.001). Post-hoc t-test analyses showed no significant interaction effects on RT and accuracy(see Supplementary Tables 3 and 4). On accuracy, the BPF group showed a statistically significant decrease compared to controls on No-Go| (p = 0.022) and No-Go+ (p = 0.006) trials. Because interaction terms did not reach significance, the following comparisons were exploratory and suggest that accuracy was sensitive to Go/No-Go task demands, indicating a possible general inhibitory/attentional deficit related to BPD features, while RT was not significantly influenced by emotional condition or BPD features.

Contrast results for prosperities of ERP components between the BPF and control groups

Given the N2 and LPP components that correlated with the BPD features, additional analyses were performed to determine whether differences in ERP amplitudes between the BPF and control groups existed. The ANOVA analysis focused on significant group effects on N2 and LPP components, with other significant main and interaction effects involving experimental conditions detailed in the Supplementary Tables 5–10. Specifically, a significant group × emotional condition interaction was observed at the left frontolateral channels for both N2 peak (F = 3.98, p = 0.021) and mean amplitudes (F = 4.48, p = 0.014). Furthermore, a significant main effect of group was evident at the same channels for N2 peak latency (F = 5.60, p = 0.020) and the medial occipital (F = 8.48, p = 0.005) and posterior lateral left channels (F = 5.68, p = 0.021). Figure 8A-C showed the group difference across conditions and Go/No-Go trials for these effects at frontolateral left channels. For LPP component, a significant interact effect of group and Go/No-Go trial type was found peak amplitude in frontolateral (F = 4.69, p = 0.033) and the medial occipital channels (F = 3.99, p = 0.020, shown in Fig. 8D and E). However, the peak latency of LPP did not exhibit a significant group effect (shown in Fig. 8F).

Fig. 8.

Prosperities of ERP components between BPF and control groups across conditions and Go and No-Go trials. Notes: This figure displays the mean peak ERP amplitudes (in µV) for the BPF (blue) and control group (green) across three emotional conditions: negative (−), neutral (I), and positive (+). Data are shown separately for Go and No-Go task conditions. Error bars represent the standard error of the mean (SEM). A-C is part of the N2 components at frontolateral left channels, while D-F is part of the LPP components at occipitalmedial channels

At the single level and post-hoc t-test analysis for N2 components, the results indicated that during the processing of negative emotional cues, the BPF group exhibited smaller N2 amplitudes in the left frontolateral region than the control, either peak (t = 2.996, p = 0.034) and mean (t = 2.938, p = 0.040). The peak latency of N2 also was latter in the BPF group than the control group across condition. For LPP components, only the control group showed a higher LPP peak and mean amplitude in Go trials compared to No-Go trials. The LPP peak amplitude of the control group significantly increased during Go trials (t = 5.491, p < 0.001), as did the mean amplitude (t = 4.418, p < 0.001). This distinction was not observed in the BPF group.

Discussion

The current research examined the relationship between specific dimensions of BPD features and ERP responses associated with emotional processing and cognitive control. The study explored how different facets of BPD, such as affective instability, negative sociality, self-harm, and identity problems, were linked to distinct patterns of neural activity during an emotional Go/No-Go task. The findings suggest that the N2 component elicited by emotional Go/No-Go cues may be associated with affective instability, while the LPP component could be linked to self-harm feature. Although the LPP component showed a relationship with identity problems, the linear association was not strongly evident. In group contrast analysis, the BPF group had smaller N2 amplitudes at left frontolateral channels during negative emotional cue processing, along with delayed N2 peak latency. LPP analysis showed a significant group-trial type interaction, with higher LPP amplitudes in Go trials for the control group. These findings suggest that there are altered neural dynamics in emotional and cognitive control for individuals of BPD features.

Affective instability and N2 component

As we found, individuals with higher affective instability show greater N2 amplitudes when trying to inhibit responses to positive stimuli; this may reflect increased neural activity related to the detection of conflict between the prepotent response and the need to inhibit that response. Our study indicates that the N2 time window, when prompted by positive-No-Go stimuli, points to an enhanced awareness of cues that signal the necessity to restrain reactions to positive stimuli. This heightened N2 component mirrors the emotion-modulated inhibitory control seen in BPD. It could signify a struggle in managing reactions to possibly gratifying stimuli, potentially appearing as impulsive actions in the pursuit of positive experiences, even if such pursuits could be harmful in the future, like making unwise financial choices [41]. This finding aligns with the idea that individuals with BPD features experience increased conflict when needing to suppress responses to positive emotional cues [26]. Hence, individuals with higher affective instability might require more neural resources to effectively monitor conflict during the task.

Specifically, the inverse correlation between affective instability and N2 components in the occipital medial region during Go|– trials (248-272ms), also within the N2 time window, presents a more nuanced picture. A reduced N2 in this context might indicate a blunted response to the conflict associated with processing negative stimuli [42]. This could suggest difficulties in effectively evaluating and responding to negative emotional cues, potentially leading to either overreaction or underreaction to negative situations. This finding aligns with research suggesting that individuals with BPD features demonstrate impairments in emotional processing, specifically in the context of negative stimuli [42]. It might contribute to the difficulties they experience in managing negative emotions, leading to heightened affective instability [43]. While in many Go/No-Go paradigms the N2 is often more pronounced on No-Go trials due to conflict monitoring and inhibitory control demands, the presence of an N2-like component during Go trials is not uncommon. However, the functional significance of the N2 during Go trials may differ from that during No-Go trials. The current findings suggest that individuals with higher affective instability show altered neural responsiveness in early conflict or attention-related processes during Go|– trials, indicating that even when a response is executed (not inhibited), individuals with affective instability have atypical processing of negative emotional stimuli. However, the differences attributed to valence-specific processes may instead point to broader distinctions in conflict monitoring or attentional allocation and highlight the complex interplay of emotion and cognition in BPD-related dysfunction.

Hence, these findings collectively suggest a complex interplay between neural processes associated with emotional responsiveness and cognitive regulation in BPD. Both of these neural patterns could contribute to the overall experience of affective instability, which is a hallmark feature of BPD [44, 45]. Furthermore, the implication of the N2 component in both positive and negative cue conditions highlights its potential role as a neurobiological marker of affective instability in BPD. Future research could explore these findings further by investigating the impact of specific emotion regulation strategies on these neural responses. Additionally, studies examining the relationship between these ERP patterns and real-world behavioral measures of impulsivity and emotional reactivity could provide valuable insights into the clinical manifestations of these neurobiological processes.

Self-harm, identity problems, and LPP component

The positive correlation between self-harm and ERP responses in the left frontal lateral region during the Go|– trials (674-704ms) points towards a potential link between self-injurious behavior and the LPP component. The LPP is typically associated with the processing of emotionally salient stimuli, and its increased amplitude in this context suggests a heightened attentional and emotional engagement with negative cues in individuals with a history of self-harm [46]. This could indicate a difficulty in disengaging from negative emotional experiences, which might contribute to the urge to self-harm as a means of regulating or escaping these intense emotions. Schmahl et al.‘s research revealed increased 400-780ms amplitudes in individuals with BPD during the processing of painful stimuli, indicating that they may require more cognitive resources to manage attention when dealing with such stimuli [47]. Given the specific association with Go|– trials, self-harm may serve to modulate emotional responses in situations requiring action or engagement with aversive stimuli. The fact that the correlation emerges in the left frontal lateral region further suggests the involvement of cognitive control and emotion regulation processes, which are often implicated in self-injurious behavior [48–50]. However, it’s important to note the relatively small R² range (0.225–0.232), indicating that while statistically significant, the LPP component explains only a modest portion of the variance in self-harm.

The results pertaining to identity problems are more complex. Although significant correlations were observed between identity problems and ERP responses in both the left frontal lateral and frontal central regions during Go|– and No-Go|+ trials, respectively, the absence of a significant linear regression effect raises questions. This suggests that the relationship between identity problems and ERP activity might not be as straightforward as initially anticipated. The positive correlations without the corresponding linear effects could indicate that other factors might be mediating or moderating this relationship. It is possible that factors like comorbid psychiatric conditions, individual differences in task performance, or other moderating variables influence this association but were not fully captured in the current models. Furthermore, the timeframes of these correlations (664-774ms for Go|– cues and 630-734ms for No-Go|+ cues) again implicate the LPP, suggesting that difficulties in processing emotionally salient stimuli, particularly in contexts requiring inhibitory control, might be related to identity problems [51, 52]. Previous studies used the Self-referential Encoding Task and found that individuals with BPD features showed significantly higher amplitudes in the LPP component (600–1200 ms) in response to negative self-referential words, whereas the control group showed higher amplitudes for positive self-referential words [53, 54]. Besides, studies ascertained that identity problems like negative thoughts about the self could be categorized as automatic thoughts, or internal ruminative thoughts which result in increased conscious rumination on negative information may be associated with enhanced LPP amplitudes [55]. While previous research has linked LPP activity to self-referential processing and emotional responses relevant to identity in BPD, our findings suggest that this relationship is complex and not always linear. Therefore, the lack of strong linear association, despite observed correlations, warrants further investigation to understand the nuanced role of neural activity in the experience of identity problems and explore the potential causal relationship with interventions.

Missing association in negative sociality and ERP components

The lack of significant associations between negative sociality and ERP components is an unexpected but interesting finding. It contrasts with studies that have previously linked social dysfunction with neural activity, especially in the context of BPD [56]. Negative sociality encompasses complex interpersonal difficulties, such as rejection sensitivity and hostile attribution biases, which involve nuanced social cognition and interactions beyond simple emotional inhibitory control captured by the Go/No-Go paradigm [57]. Instead, tasks specifically designed to engage social feedback, exclusion (e.g., Cyberball task), or trust-related decision-making would provide a more direct assessment of these interpersonal dimensions. Incorporating such paradigms could yield stronger and more relevant ERP-feature associations for negative sociality. For example, cyberball is a virtual ball-tossing game used to simulate social exclusion in experiments. During the game, participants are led to believe they are playing with other individuals but are eventually excluded from the ball tossing. This paradigm allows researchers to examine neural responses to social rejection, with key ERP measures including the feedback-related negativity (FRN) and LPP, both of which may be enhanced in individuals with BPD due to heightened sensitivity to rejection [58, 59].

In the present study, a significant P3 component was not observed. While P3 differences might exist within single conditions, these contrasts may not effectively explain the BPD traits under investigation. One possible explanation for the lack of a significant P3 component could be related to the task design or the specific cognitive processes engaged by individuals with BPD features during the emotional Go/No-Go task. P3 component is highly sensitive to task demands and individual differences in attention allocation. Therefore, the absence of a significant P3 component might also reflect variations in attentional strategies or cognitive resource allocation among participants with BPD. Future research should focus on refining experimental paradigms to better capture the neural mechanisms in BPD.

A synopsis of group-based disparities in ERP components

The study reveals significant differences in ERP components between the BPF and control groups, particularly concerning the N2 and LPP. The BPF group displayed smaller N2 amplitudes during the processing of negative emotional cues and a later N2 peak latency in the left frontal electrodes, indicating reduced and delayed conflict monitoring. Furthermore, the LPP findings suggest a context-dependent difference in emotional processing between groups, with the control group showing elevated LPP during Go trials (as expected) and the BPF group failing to exhibit this distinction. The results are in line with prior studies that have pointed out the struggles in merging cognitive control with emotional reactivity in individuals diagnosed with BPD [24]. Notably, these differences are particularly pronounced during the processing of negative cues aligns with the idea that individuals diagnosed with BPD often show heightened sensitivity and reactivity to negative emotional information [26]. It is important to clarify that where post-hoc analyses indicated differences for specific cues or emotions, these findings represent potential trends rather than definitive cue- or emotion-specific effects. To confirm such specificity, future research with larger sample sizes or study designs capable of detecting interactions is warranted.

Additionally, although ERP differences were observed between groups, these neural variations did not always correspond to measurable behavioral changes, suggesting they may reflect latent neural processing differences rather than overt functional impairments. It emphasizes the importance of utilizing a multi-modal approach, combining sensitive neurophysiological measures like ERPs with a broader range of behavioral assessments (including ecologically valid tasks, daily diary methods, or clinical interviews) that are specifically designed to capture the functional implications of these latent neural processing differences. This will allow for a more complete understanding of how these neural variations contribute to the overt symptoms and impairments experienced by individuals with BPD. Moreover, longitudinal investigations are crucial for a comprehensive understanding of the dynamic interplay between neural activity, emotional processing, and BPD characteristics, as suggested by the observed N2 and LPP differences.

Limitations and future research directions

The study acknowledges several limitations that should be considered when interpreting the findings. First, our analysis set a criterion of retaining a minimum of 50 samples in ERP components for data processing, which may have influenced the event-related potential amplitudes observed in subsequent analyses. This serious criterion might not have completely revealed the relationship between them. Especially, the squared association may be neglected in the small continued samples. Second, as for statistical concerns, the R² values were generally modest, and associations were described in cue-specific terms without showing significant differences from other cue types. More factors may influence the relationship between ERP components and specific BPD features. Future research should explore and incorporate other potential mediating variables for a more comprehensive understanding of the neural mechanisms underlying BPD. Besides, these preliminary findings indicate a potential connection between ERP components during emotional inhibitory control and certain BPD feature dimensions. Strong correlations between PHQ, GAD, and PAI-BOR scores suggest that some group/feature associations may be due to general distress rather than BPD-specific inhibitory deficits. This substantial overlap suggests that some of the observed associations between BPD features and inhibitory control deficits might partly reflect general distress or comorbid symptomatology rather than BPD-specific mechanisms. Accordingly, it is challenging to fully disentangle unique effects attributable solely to BPD-related emotional dysregulation from those associated with broader psychopathology dimensions. Future research employing approaches such as statistical controls, longitudinal designs, or more refined clinical phenotyping will be important to clarify the specificity of inhibitory control impairments in BPD. Lastly, it is important to note that while the PAI-BOR is a useful screening tool for quickly assessing an individual’s impulsivity and BPD-related features, there are limitations to consider. Self-reports depend on participants’ honesty and self-awareness, making them susceptible to biases such as social desirability bias and memory biases that can affect the accuracy of the results. In order to improve the accuracy and practicality of upcoming studies, we suggest including structured clinical interviews to verify BPD diagnoses instead of relying solely on self-report screenings. Furthermore, no validated Chinese version of the PAI-BOR currently exists and cultural differences in personality expression warrant caution. Thus, developing and validating a culturally adapted Chinese version of the PAI-BOR should be a priority for future research.

Conclusion

This study provides insights into the neural correlates of emotion-modulated inhibitory control in relation to specific features of BPD. The observed patterns of event-related potential activity during an emotional Go/No-Go task suggest that the N2 component, which reflects emotional processing and response inhibition, may be associated with the core feature of BPD, such as affective instability. Specifically, the tendency for self-harm correlates with increased LPP during the processing of negative cues, while identity problems are slightly linked to altered LPP components. The unexpected lack of association between negative sociality and ERP components highlights the complexity of social information processing in BPD. This research identifies specific neural correlates of emotion-modulated inhibitory control associated with the core feature of BPD, contributing to a more comprehensive understanding of the underlying mechanisms driving the emotional and behavioral difficulties experienced by individuals with BPD features.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BPD

Borderline Personality Disorder

PAI-BOR

The Personality Assessment Inventory - Borderline Features

PHQ-9

Patient Health Questionnaire-9

GAD-7

Generalized Anxiety Disorder-7

ERP

Event-Related Potential

ROIs

Regions of Interest

R2

R-squared

LPP

Late Positive Potential

Author contributions

Yin Qianlan: conceptualization, methodology, data curation, writing – original draft, writing – reviewing & editing. Shu Tong: Project administration, writing – original draft; writing – reviewing & editing. Chen Zhuyu: Writing – original draft, writing – review & editing.Xu Huijing, Jiang Qian, Meng Liang,: investigation. Liu Taosheng: Funding acquisition; writing – reviewing & editing.

Funding

This work was supported by the Key Projects in the 14th Five-Year Plan for Cyberspace Information [145AWX200020011X]; the Project of prevention and protection for mental health [ (2024)432 − 73]; and the SMMU Teaching Case Library Construction Project [ZD2024005].

Data availability

The dataset (s) supporting the conclusions of this article is (are) available in https://accounts.osf.io/login?service=https%3A%2F%2Fosf.io%2Fuw2s4%2F%3Fview%5C_only%3D7286b6067dd2454b8c350dc2f7252947.

Declarations

Ethics approval and consent to participate

This study was reviewed and approved by the Ethics Committee of Naval Medical University, with the approval number: [NMUMEREC-2021-043]. All participants provided informed consent to participate in the study.

Consent for publication

Written informed consent to include de-identified information in any subsequent publications was obtained from all participants at the time of data collection.

Competing interests

The authors declare no competing interests.