Healthy participants (N = 52; 29 women) aged from 18 to 55 years (M = 26.67, SD = 6.92) were recruited through public advertisements and flyers. All participants were required to be right-handed, have normal or corrected-to-normal vision and fulfill MR-scanning criteria. To ensure that all questionnaires could be answered, fluency in German was required. Participants with acute or history of neurological illnesses, substance-related or psychiatric disorders were excluded.

Due to poor EEG signal quality, one participant had to be excluded from the EEG data analysis reducing the final sample to 51 participants (28 women) with a mean age of 26.73 years (SD = 6.98).

The study protocol was conducted in accordance with the Declaration of Helsinki (World Medical Association, 2013) and was approved by the Ethics committee of the RWTH Aachen Medical Faculty. All participants gave written informed consent and received a financial compensation of 45 Euro.

Data collection took place over a period of 18 months at the Department of Psychiatry, Psychotherapy, Psychosomatics, Medical Faculty RWTH Aachen University.

Included participants were invited to a 3.5-h measurement appointment. First, participants completed questionnaires. A 64-channel EEG system (BrainAmp MR Amplifier, Brain Products GmbH, Gilching, Germany; BrainAmp [Apparatus], 2023) with an MR-compatible electrode cap was placed on participants’ heads. The 64 electrodes (5 kΩ resistors) included an electrocardiogram electrode (ECG) placed on the back left to the spine, the vertex electrode (Cz) on the middle of the scalp was used as reference electrode.

For the simultaneous EEG-fMRI measurement, participants were positioned in a 3 Tesla MR scanner (Siemens MAGNETOM Prisma®, Siemens Medical Systems, Germany) while their right hand was placed on an MRI compatible button device. For synchronization of EEG and MRI recording a synchronizing box (SyncBox MainUnit, BrainProducts, Germany) was used. To stabilize the head position, a cervical collar was placed.

The measurement session started with performing anatomical brain measurements. After completing two tasks in the scanner participants were asked to fill in a task evaluation questionnaire.

EEG data were recorded with the BrainVision Recorder software (BrainVision Recorder, Vers. 1.23.0003, Brain Products GmbH, Gilching, Germany; BrainVision Recorder (Vers. 1.23.0003) [Software], 2023) at a 1,000 Hz sampling rate (0.01–250 Hz analog band-pass filter). The BrainVision Analyzer software (BrainVision Analyzer, Version 2.2.0, Brain Products GmbH, Gilching, Germany; BrainVision Analyzer (Version 2.2.0) [Software], 2019) was used for EEG data preprocessing. The simple gradient method was used to remove MR scanner artifacts (Allen et al., 2000). An infinite impulse response filter (IIR, 70 Hz cut off, 48 dB slope) was applied and data were down-sampled to 500 Hz which acts as a low-pass-filter. Data then underwent cardioballistic pulse artifacts detection and correction (Allen et al., 1998). Linear topographic interpolation was performed on EEG channels selected after visual inspection. After down-sampling to 250 Hz, data were filtered (0.01–45 Hz, 24 dB slope; ECG channel excluded from filter). Independent component analysis (ICA) was used to identify, e.g., eye blink artifacts which were removed from the data. All datasets were visually inspected, and intervals of bad quality were semi automatically detected and rejected. EEG data were exported for further analyses via EEGLAB (MATLAB®).

Further EEG data processing was done by using MATLAB software (version 9.9.0.1467703 (R2020b). Natick, Massachusetts: The MathWorks Inc, 2020) and the implemented toolbox EEGLab (version 2022.1; Delorme and Makeig, 2004). The preprocessed EEG data were segmented to epochs ranging from −200 to 800 ms relative to the primer onset (Nguyen and Cunnington, 2014). For target analysis, epochs were expanded to a range from −200 to 1,200 ms relative to primer onset. Baseline correction for the interval of 200 ms prior to the primer onset was carried out (Nguyen and Cunnington, 2014; Zhang et al., 2016).

For ERPs analysis referring to the primer (P100 and N170), the channels P7, PO7, P8 and PO8 were chosen (Nguyen and Cunnington, 2014). Mean ERPs of the channels P7 and PO7 (left hemisphere) as well as the channels P8 and PO8 (right hemisphere) were created for every participant. For the primer analysis, P100 and N170 were computed for every subject by averaging the segments for each primer condition (happy, sad, and neutral primer emotion × unconscious and conscious primer presentation; Figure 2).

The N170 peak component (negative) was estimated between 152 to 200 ms after primer stimulus onset (Hinojosa et al., 2015). Following previous studies, the P100 time interval was initially set to 80 to 120 ms (Nguyen and Cunnington, 2014). After visual data inspection, the P100 peak component (positive) was estimated in an extended time interval ranging from 80 to 132 ms after primer stimulus onset.

The identified negative (N170) and positive (P100) peaks within the time windows were subjected to statistical analysis using R (R Core Team, 2022) and RStudio (version 2022.12.0.353; Posit team, 2022).

No statistically significant differences regarding age, and questionnaire scores were found between men and women (Table 1).

For an exploratory analysis of effects of different questionnaire scores on the ERP components as well as participants’ accuracy, see Supplementary material.

A significant effect of congruence was found [F(1,863) = 4.74, p < 0.001, Cohen’s f = 1.93; for all results see Table 5]. Response accuracy was greater in congruent than in incongruent trials (Figure 7). Post hoc comparisons indicated significant differences between all three target emotions [happy-neutral t(863) = 6.31, p < 0.001; happy-sad t(863) = 17.92, p < 0.001; neutral-sad t(863) = 11.63, p < 0.001; Table 5]. Participants’ responses were most accurate in happy target faces (happy > neutral > sad; Figure 7). The model explained 10.47% of variance via the fixed factors and 79.46% via the random intercept.

The result of larger P100 amplitudes in unconscious than conscious trials is in line with past research. For example, larger P100 amplitudes were reported in trials where facial stimuli were presented for 17 ms (unconscious) in comparison to 200 ms (conscious) presentation time (Zhang et al., 2012). Assuming a modulation of P100 amplitudes by attention as reported in some literature showing higher P100 in trials with increased attention (Mangun and Hillyard, 1991; Mangun, 1995), briefly presented stimuli may elicit higher attention than trials with longer presentation times. This may lead to higher P100 amplitudes due to the increased attentional demand (Hillyard et al., 1998). A review proposed that some unconsciously perceived emotional facial expressions may attract higher attention, aiming to gain awareness of the emotional face (Eastwood and Smilek, 2005). Therefore, attention effects may be highly relevant especially for shortly presented facial expressions and are reflected by P100. Additionally, unconsciously perceived stimuli may have been processed only partly at this early stage which could increase the attentional demand resulting in higher P100 values.

N170 was larger in conscious compared to unconscious trials. The N170 is sensitive to face stimuli (Hinojosa et al., 2015) and fixation on specific face parts, e.g., on eyes elicited larger N170 amplitudes than fixation on mouths (de Lissa et al., 2014). Therefore, trials with longer stimuli presentation times may enable more detailed face processing resulting in larger N170. Our findings are in line with previous research proposing sensitivity of the N170 component to the visibility of faces (Zhang et al., 2012). Our results demonstrate a clear modulation of P100 and N170 amplitudes by stimulus presentation time, reflecting differential neural processing in conscious versus unconscious perception. The larger P100 amplitudes in unconscious trials suggest an early, rapid attentional mechanism that is activated by brief, unconsciously perceived stimuli. This heightened P100 response may indicate that the visual system allocates more attentional resources to quickly process stimuli that are only partially perceived, aligning with the hypothesis that unconscious emotional stimuli can attract higher attention (Eastwood and Smilek, 2005). On the other hand, the N170 amplitude was larger in conscious trials, consistent with the notion that conscious face perception engages more detailed and prolonged visual analysis, allowing for full engagement of face-specific cortical regions such as the fusiform gyrus (Zhang et al., 2012). This ERP has found to be altered depression on an unconscious level as well (Zhang et al., 2016), showing implications on how this ERP can be used to further understand the mechanisms of mental disorders. Furthermore, N170 and P100 can be used to classify unconscious vs. conscious processing and could be used to study more real life-like situations including subliminal facial expression processing using, e.g., dynamic stimulus material and can be therefore adapted to elaborate on disease specific alterations in unconscious emotion processing. Unconscious processing has practical relevance in daily interactions, where subtle emotional cues in faces may be registered unconsciously, influencing reactions before conscious awareness. This type of rapid processing could assist in social situations by helping to assess emotions or detect threats without requiring full awareness. Neuroscientific evidence supports the biological plausibility of processing complex stimuli like faces without awareness. Studies indicate that brain regions like the amygdala and subcortical pathways respond to faces even when not consciously perceived, suggesting an evolutionary adaptation for rapid, unconscious assessment of socially relevant cues (Pessoa, 2005; Tamietto and de Gelder, 2010). These findings align with our study’s aim to explore unconscious processing, showing its importance beyond controlled settings.

We hypothesized that the P100 component might be larger in trials displaying happy or sad faces compared to trials with neutral faces. However, the P100 was not significantly affected by the presented emotion in our study which could be due to a small effect size or the power of our model including the factor “emotion.” Since we observed wide confidence intervals that indicate a large data variability, we assume different sources that are not captured in our model. With a larger sample size, we could enhance precision about the effect and report more reliably reproducible effects. Other studies did not find an emotion effect on P100, as well (Batty and Taylor, 2003; Smith, 2012; Lee et al., 2017). The deviating results may be explained by the use of different methods calculating the ERP amplitudes, averaging the amplitude in a given time window (Zhang et al., 2016), contrary to selecting the peak within the time window. Another possible explanation may be differences regarding the experimental design, e.g., the longer stimulus presentation time of 500 ms, or the absence of a mask (Moradi et al., 2017).

As indicated by previous results (Hinojosa et al., 2015; Batty and Taylor, 2003; Miyoshi et al., 2004; Blau et al., 2007; Hendriks et al., 2007; Pegna et al., 2011; Smith, 2012; Zhang et al., 2012, 2016), the N170 component was influenced by different emotions. While the main effect of emotion on N170 was significant, post hoc comparisons between primer emotions were not significant. Since confidence intervals were wide and included the null value, large data variability might be considered as an influencing factor on the results. A meta-analysis underlines the effect of the emotion ‘happy’ on the N170 amplitude compared to neutral faces, whereas it does not substantiate the effect of the emotion ‘sad’ compared to neutral faces (Hinojosa et al., 2015).

Despite the robust effect of happy emotions, there are also contradictory findings with larger N170 amplitudes in sad faces than in happy or neutral faces (Lynn and Salisbury, 2008). There might be interindividual differences and preferences towards a specific emotion facilitating its processing. Assuming that such a bias towards specific emotions exists, this would result in increased ERPs for a specific emotion. In our study, the results would not support a specific bias towards one specific emotion, as no significant difference comparing happy and sad emotional faces was found. However, others found biases towards emotional information (for review see Kauschke et al., 2019).

We found a positive correlation between the N170 and subjective difficulty of evaluating emotion within the task paradigm. Participants who reported emotion assessment to be easier, elicited larger N170 amplitudes. Therefore, the emotion effect on the N170 may be dependent on participants’ general sensitivity and ability to discriminate emotions.

Early ERPs were found to be associated with lower-level processing, whereas later ERPs reflected more complex cognitive processes (Sur and Sinha, 2009; Portella et al., 2012). Even though the temporal distance between P100 and N170 is short, this is a substantial duration in the context of brain processing. This trend may be reflected by the finding that the N170 amplitude was modulated by presentation time as well as emotion, while the earlier P100 component was only influenced by stimulus presentation time.

The present study found significantly larger P100 amplitudes in women compared to men, whereas the N170 showed opposite results. Previous work investigating the effect of sex on ERPs related to emotional face processing found that women elicited larger P100 amplitudes than men in subthreshold fearful faces (Lee et al., 2017). Using an oddball paradigm, one study found that women elicited larger N170 amplitudes in emotional compared to neutral faces, whereas no such effect was found in men (Choi et al., 2015). Since previous study designs differed in experimental paradigm and emotions presented, it is unclear whether these findings are applicable to discuss our results. As the amount of research conducted on the effect of sex on the P100 and the N170 component is still insufficient, further research is needed for conclusive results. Additionally, it should be considered that the confidence intervals were wide indicating larger data variability and included the null value highlighting the need for more extensive research on the influence of the factor “sex” on the ERP components.

Both the P100 and the N170 amplitudes were larger on the right compared to the left hemisphere. In accordance with this, past findings suggest right hemisphere dominance for face processing in the N170, whereas results on the P100 are not fully conclusive yet (e.g., Leehey et al., 1978; Levine et al., 1988; Bötzel et al., 1995; Bentin et al., 1996; Kanwisher et al., 1997; Haxby et al., 1999; Rossion et al., 2003; Itier and Taylor, 2004; Zhang et al., 2012; Nguyen and Cunnington, 2014).

The N170 was increased with higher age which is in line with previous studies which discovered a trend towards an increase in N170 amplitudes in older participants. However, these studies were mainly focused on the population of children and adolescents (e.g., Taylor et al., 1999; for a review see Taylor et al., 2004). One study reported no significant difference of N170 amplitudes between young and middle-aged participants (Shi et al., 2022). Considering the face sensitivity of the N170 (Bentin et al., 1996; Hinojosa et al., 2015), it could be hypothesized that face processing still develops and matures in adults over the course of life. Accounting for the width and inclusion of the null effect in the confidence intervals analyzing the effect of age on the N170, our results should only be applied cautionary to other populations and more research on the effect of age in adult populations on the amplitude of the P100 and the N170 is required to establish conclusive results.