Emotional information carries biological and social significance in our daily lives from when we attempt to avoid a life-threatening danger to when we try to understand others. How we process the emotional information is closely related to how we interpret other's intentions and how we behave. Moreover, identical emotional information could affect each of us differently. Individual difference in processing emotion could explain the mechanism of emotion processing. There are several factors known to affect the emotion processing such as personality traits, experiences, or genetics (Hamann and Canli, 2004). For example, extravert personality trait showed positive correlation with the degree of activation in the amygdala, which is known to be involved in emotion processing, when happy faces were viewed (Canli et al., 2002). Activity in the amygdala was also correlated positively with pessimism while unpleasant stimuli such as snakes were viewed (Fischer et al., 2001). In another study, differences in voluntary emotional regulation across individuals evoked by emotional visual stimuli modulated activation in the amygdala (Schaefer et al., 2002). Moreover, people with a genotype such as 5-HTT showed enhanced activity in the amygdala when they view faces with emotional expressions (Hariri et al., 2002).

Gender is another source of individual differences in response to emotional information (Bradley et al., 2001; Kret and De Gelder, 2012). Several studies have reported gender differences in trait and neural responses to emotional stimuli. Women were found to be more expressive as shown by more responsive electromyography (EMG) to emotional face stimuli (Dimberg and Lundquist, 1990) and greater skin conductance response (SCR) to emotional films (Kring and Gordon, 1998). Several studies have demonstrated that women show higher sensitivity in recognizing emotional information presented not only in visual modality (Kring and Gordon, 1998; Montagne et al., 2005; Collignon et al., 2010) but also in auditory modality (Collignon et al., 2010). Women's sensitivity to emotional information is further supported by higher accuracy in emotional recognition and categorization (Thayer and Johnsen, 2000; Hall and Matsumoto, 2004). In response to emotional stimuli, women tend to show facial expressions more accurately labeling the emotional contents (Wagner et al., 1986), and different patterns of SCR (Kring and Gordon, 1998) and startle reflex (Bianchin and Angrilli, 2012) to emotional stimuli compared with men.

Some studies have reported that this sensitivity to emotional information in women may explain why women are more likely to suffer from psychiatric problems than men (McGrath et al., 1990; Nolen-Hoeksema and Girgus, 1994). There are many studies that found women to be more sensitive and responsive to unpleasant stimuli such as sad faces while men are more sensitive to stimuli with positive valence (Williams and Gordon, 2007; Li et al., 2008). Gasbarri et al. (2007) have reported that women are more selectively attentive to biologically relevant stimuli with negative emotional valence such as sadness and fear. Bradley et al. (2001) have found that women react more defensively as observed by unpleasant facial muscle activity (EMG), cardiac deceleration, and startle reflex to unpleasant stimuli, which all reflect intensified selective attention. Lang et al. (1998) observed enhanced activation in the occipital cortex in response to unpleasant stimuli in women and pleasant stimuli in men.

Studies using event-related potentials (ERPs) have reported gender differences in neural responses to perceptually visible emotional stimuli, which is reflected in enhanced activity in women compared with activity in men. The gender-related ERP differences have been observed in both early and late responses. For example, N200 component was enhanced more when women viewed unpleasant stimuli compared to when men viewed the same stimuli (Lithari et al., 2010). Campanella et al. (2004) found using the oddball paradigm that N2b latency differentiated happy and fearful stimuli in men while such differentiation was not found in women. In another study, women showed more enhanced P2, the component implicated in higher-order attentional processing, compared to men in response to unanticipated negative stimuli in a modified cue-target paradigm (Jin et al., 2013). These results all support that women have attentive bias to emotional stimuli, particularly to unpleasant stimuli.

Enhanced responses in late components such as P300 and LPP are also observed in women (Oliver-Rodríguez et al., 1999; Gasbarri et al., 2006, 2007; Han et al., 2008). Gasbarri et al. (2006) observed that women showed larger P300 responses and better memory retrieval for emotional stimuli than men. Difference in even later component, P450, was observed in Orozco and Ehlers's (1998) study in which women showed overall longer latency and higher amplitude in response to both happy and sad faces. In another study by Luo et al. (2014), only women showed longer latency of late positive potential (LPP) to moderate negative stimuli drawn from International Affective Picture System (IAPS) despite no gender difference in early responses.

Taken together, these studies suggest that women show higher sensitivity and more responsiveness when they view emotional stimuli of negative valence, which might be the source of the tendency that women are more vulnerable to affective disorders such as anxiety and depression compared to men (Nolen-Hoeksema, 1987). It is noteworthy that in most of the previous studies, emotional information was delivered to the participants with clear visibility by presenting the stimuli for long duration with full contrast. However, emotional information can also be fleeting and feeble. A considerable amount of evidence has shown that emotional information can be processed without conscious awareness, indicative of behavioral and neurophysiological responses to perceptually invisible emotional stimuli (Tamietto and de Gelder, 2010). However, few studies have investigated whether men and women differ in their response to emotional stimuli. Among those few studies, Hall and Matsumoto (2004) showed women's superior ability recognizing emotional facial expressions even when the stimuli were presented so fast as to be “at the edge of conscious awareness.” Specifically, they presented each face in one of six expressions (anger, contempt, disgust, fear, happiness, sadness, or surprise) briefly (70, 130, or 200 ms) in the middle of 1-s presentation of a neural face of the same person. As in the prolonged viewing condition, women were more accurate than men recognizing each briefly presented expression. More recently, Donges et al. (2012) employed a version of subliminal affective priming procedure where the valence of a subliminally processed emotional facial expression influences the valence of a subsequent target to examine gender difference in response to invisible emotional stimuli. The results showed that despite the lack of conscious awareness of the briefly presented and masked happy prime face (30 ms), women judged the subsequent neutral face more positively compared to men. Interestingly, such gender difference in subliminal affective priming was not significant in the case of sad prime face. Based on those limited number of recent studies, it is important to investigate the gender differences in neurophysiological responses to perceptually invisible emotional stimuli, to further support the association between women's vulnerability to affective disorder and the gender difference in emotion processing below conscious awareness.

In the current work, we set out to examine whether there exists gender differences in response to emotional stimuli processed outside conscious awareness by rendering the stimuli perceptually invisible. For that, we used a backward masking paradigm to render a fearful or neutral face invisible while measuring ERP responses to it. The duration of the masked face was varied, which enabled us to compare ERP responses to the subthreshold emotional stimuli with ERP responses to stimuli processed at near-threshold and suprathreshold. ERP components of interest included P100, N170, early posterior negativity (EPN), N250, and P300, which have been implicated in affective face processing in a range of stages. We also took other personality factors into consideration by including a battery of questionnaires such as State and Trait Anxiety Inventory-trait (STAI-trait; Spielberger, 1970), Positive and Negative Affect Schedule (PANAS; Watson et al., 1988), Beck's Anxiety Inventory (BAI; Beck et al., 1988a), and Beck's Depression Inventory (BDI; Beck et al., 1988b). To disentangle neuroticism factor from the gender differences, we performed an additional, control experiment by including participants whose BAI and BDI scores are within healthy ranges. This consideration was particularly important due to several reasons; First, gender difference in emotion processing may be partly related to the higher incidence of neuroticism in women than in men (Kessler et al., 2005). Second, women's relative vulnerability to neuroticism may be related to involuntary and automatic bias toward emotions (Mayer and Merckelbach, 1999). Last but not least, a recent study showed the interaction between gender and neuroticism factors in disengaging attention from the location of invisible fearful faces (Tan et al., 2011). In the main and the control experiments, we hypothesized that the early response to fearful faces processed at subthreshold would be enhanced for women compared to men as observed in P100, EPN, or N250 components, which are relevant to low-level perceptual processes. The later response to consciously, or fearful faces processed at suprathreshold would be enhanced for women compared to men as observed in P300 or N170 components, relevant to higher-level perceptual processes of emotional faces.

In the main experiment, we have shown the gender difference of the early ERP responses to emotional face stimuli at the posterior electrode sites. Specifically, women showed larger amplitude of P100 component at O1 (Figure 4B) and O2 (Figure 4C) compared to men, which was only significant when they were presented with fearful faces too briefly so that it was processed at subthreshold. Such gender-related difference was not shown in the behavioral responses, in which both men and women showed a chance-level emotion discrimination performance. In an additional control experiment, with replacement of some participants whose BAI and BDI scores were beyond the healthy range, we replicated the gender difference in P100 amplitude in response to subthreshold fearful faces. Therefore, the larger P100 amplitude in women's posterior electrode sites than men's in response to emotional information processed at subthreshold was proven to be gender-specific, not mainly driven by clinically anxious or depressive personality factors.

The P100 ERP component is commonly thought to reflect low-level visual feature processing and be generated in early visual areas (Heinze et al., 1994). Therefore, the current results might be interpreted as gender differences in low-level visual processing in general. However, P100 has also been implicated in early face-specific processing (Itier and Taylor, 2004) even when low-level stimulus feature was controlled (Herrmann et al., 2005). Successful categorization of visual stimuli as faces was found to correlate with the early P100 component (Clark et al., 1996; Linkenkaer-Hansen et al., 1998; Itier and Taylor, 2002, 2004). The P100 component has also been implicated in emotion processing including correct detection of visual facial expressions (Utama et al., 2009). Amplitude of the P100 component is modulated by auditory emotional stimulus accompanied by visual emotional stimulus (Gerdes et al., 2014). Considering that the difference of the amplitude of P100 between men and women was observed only when the participants viewed fearful faces at subthreshold, our results are likely to reflect gender differences in visual processing of emotional information in faces.

We are not the first to report such gender-related P100 difference in emotional processing. Sass et al. (2010) have shown that a group of participants with self-reported high anxious arousal showed larger P100 than healthy controls in response to emotionally arousing words while the participants had to name the ink color of the word. Of more relevance to our current interest, women of high anxious arousal showed greater P100 than did men of high anxious arousal in response to emotional words including both threatening and pleasant ones. What makes the current results distinct from Sass et al. (2010) is the gender difference in P100 in response to emotional stimuli was observed when the stimuli were invisible.

In the current study, the gender difference in P100 responses was specific to the subthreshold fearful faces, not to the fearful faces viewed for longer durations and processed further. P100 has indeed been implicated in the subthreshold face processing in some previous studies. For example, Saito et al. (2007) have shown that the P100 amplitude in the occipital electrodes in response to faces was different significantly from the P100 amplitude in response to non-face stimuli when the stimuli were presented for 20 ms and processed outside conscious visual awareness. Moreover, the P100 amplitude was smaller for inverted faces than for upright faces when the face stimuli were processed at subthreshold duration. Such face-specialized P100 responses were not observed when the faces were presented for 30, or 300 ms, which were around and above the temporal threshold levels. These results suggest unconscious, face-specific processing reflected in the early P100 ERP component. Our results further extend the previous findings by showing P100's involvement in the unconscious emotion-specific face processing, which distinguishes women from men. Namely, the early automatic response to emotional information is amplified in women.

Individual differences in subthreshold emotional processing have also been addressed previously. A majority of those studies focused on neurological factor such as anxiety as the main source of individual differences, rather than gender, the focus of the current work. In one study, Li et al. (2007) tracked P100 responses of non-patient participants based on the degree of trait anxiety indicated by bispectral index (BIS) scores. Results showed a positive correlation between P100 amplitude associated with invisible emotional word and the trait anxiety. Specifically, the difference between P100 amplitude in response to threat-related words of subthreshold duration and P100 amplitude in response to neutral words of subthreshold duration was greater as the BIS score increased. Yet, in another study, the correlation was in the opposite direction; Walentowska and Wronka (2012) found a negative correlation between P100 amplitude and anxiety, implying that decreased early, and automatic sensitivity to subthreshold face in high anxious group. It is not our main concern to test these seemly contradictory results in previous studies. However, our results are suggestive of the apparent contradiction. There exist greater prevalence rates of anxiety disorders in women than in men (Cahill, 2006). Namely, our results showing the amplified P100 response to subthreshold fearful faces in women provide indirect evidence supporting the results of Li et al. (2007) reporting the positive relationship between P100 amplitude associated with invisible emotional word and the trait anxiety.

The tight coupling between gender and neuroticism factors is what requires great caution. A host of studies have shown the distinctive early stage of emotional processing exemplified by P100 in patients with schizophrenia (Javitt et al., 1993; Foxe et al., 2001; Campanella et al., 2012) and depression (Fotiou et al., 2003). This is of concern for most of the previous studies on gender differences in emotional processing, since other neuroticism factors might have covaried with gender. In a previous study, for example, N2 and P3 modulation only in women disappeared when depression, anxiety and alexithymia scores were matched between men and women in a modified emotional oddball task (Campanella et al., 2012). Further analysis has revealed that personality factors such as alexithymia were better at predicting the N2 latencies than the gender factor. It is also plausible that some personality factors lie at the core of what really distinguish men and women. Indeed, the main experiment of the current study showed statistically significant differences in STAI-trait and BAI, and BDI scores between men and women, despite the consideration of scores from the anxiety and depression questionnaires as covariates in the analyses. To disentangle these two sources of individual differences, most of the previous studies have attempted to show that patients with emotional disorders had automatic attentional bias toward negative emotions compared to healthy controls regardless of gender of the patients. For instance, patients with anxiety or depression had pre-attentive processing bias for anxiety- or depression-related words compared to healthy individuals regardless of the gender (Mogg et al., 1993). In contrast, the current study tried to tease the neuroticism factor from out of the gender factor by replacing the participants whose scores in those measures are higher than the healthy ranges with the new participants whose scores are within the healthy ranges in the depression control experiment. Since the greater P100 in women than in men in response to invisible emotional faces was replicated in the depression control experiment, we are more confident that the P100 difference between men and women is not entirely driven by the neuroticism factors. This last point is what makes the current work stand out amongst other previous works.

It should be noted, however, that the questionnaire scores, BAI and STAI-trait in particular, differed between men and women even when the acceptable range of anxiety and depression was controlled. Although the difference may have still affected the neurophysiological response even after it was controlled as a covariant, we believed that discarding the difference might incur rather an unnatural phenomenon. It is often accepted that men and women differ in emotional balance even in healthy population. The inherent gender difference in anxiety and depression level may be the natural phenomenon, therefore we did not try to artificially match scores for men and women. The current study demonstrated that there might be a biological difference between men and women in processing emotional information. However, relevant personality factors such as anxiety or depression could also be significant factors that cannot be separately considered. Future study should consider such personality factors when investigating gender difference in automatic response bias to emotional information.

In addition to P100, we found a few more significant results in terms of other ERP components.

First, there was hemisphere lateralization in N170 component regardless of awareness. N170 amplitude to fearful face was significantly larger at left hemisphere compared to right hemisphere only in women. This result is consistent with previous studies that observed dominance of left hemisphere to unpleasant stimuli in women and dominance of right hemisphere to pleasant stimuli in men. Several studies have observed gender difference of hemisphere lateralization at N170 component (Proverbio et al., 2006). A number of studies have shown right hemisphere dominance at N170 component to face stimuli especially when there are more men among the participants (Campanella et al., 2000; Itier and Taylor, 2004; Harris et al., 2005; Kovács et al., 2006). When there are more women among the participants, N170 activity seems to be either bilateral or more dominant at left hemisphere (Jemel et al., 2005; Meeren et al., 2005; Pourtois et al., 2005). In Righart and de Gelder study (2006), 10 among the 12 participants were women. They observed larger N170 amplitude to faces in fearful context compared to that in neutral context in the left hemisphere. Hemisphere lateralization is also observed in other components and brain areas. For instance, Gasbarri et al. (2007) observed enhanced P300 amplitude and latency to unpleasant pictures at left hemisphere for women while men showed enhanced P300 at right hemisphere. Although we did not find hemisphere lateralization for other components, it seems notable that only women showed left hemisphere dominance for face-specific component, which further indicates different brain mechanism of emotion processing between men and women.

Second, longer N250 latency to subthreshold compared to suprathreshold condition was observed in the current study. Several studies have suggested the correlation between task difficulty and longer N250 latency (Towey et al., 1980; Letourneau and Mitchell, 2008). To discriminate invisible facial expressions may have made decision-making more difficult for the participants resulting in longer N250 latency. However, it should be noted that we did not observe enhanced N250 amplitude or P300 amplitude to subthreshold fearful face compared to suprathrehsold fearful face. Previous studies have observed modulation of N250 and P300 complex by visibility of stimuli (Liddell et al., 2004; Williams et al., 2004; Kiss and Eimer, 2008). Although the awareness effect for these components was not our main interest, it may be taken into consideration that we did not find awareness effect for frequently observed components. The absence of differential activity N250 and P300 should be further investigated using different paradigms of making stimuli invisible.

The current study attempted to investigate the gender difference in processing emotional information at various levels of awareness. There exists a host of evidence that women are more responsive and expressive to emotional information compared to men, but the results from the current study demonstrated that healthy women have more intensive automatic response to negative emotional information at subthreshold level compared to men. This gender difference at early level of emotion processing (enhanced P100 amplitude to subthreshold fearful face in women) was observed even when depression and anxiety were controlled, implying further that women's vulnerability to the threats of depression and anxiety may be related to the different neurophysiological responses. Furthermore, faster and enhanced early response to subliminal emotional information in women is in line with how women tend to be more sensitive to stimulus in weaker intensity. Our results suggest that women's sensitivity to emotional stimuli which has been developed though evolution and social learning has biological bases. Even though there are some shortcomings in our study, such as the absence of gender difference at suprathreshold condition, this appears to be the first attempt to investigate gender difference of processing emotional information at subthreshold level of awareness in healthy subjects. To identify that this observation is not limited to a specific paradigm, a further study is needed such as using different paradigms inducing stimuli perceptually invisible.

SAL suggested the idea, conducted the experiment, analyzed the results, and wrote the whole manuscript. MS helped the ERP analysis. SHL and CK designed the study, and edited the manuscript.

This research was supported by the Brain Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning (NRF-2015M3C7A1028252). This research was also supported by the NRF funded by the Ministry of Education, Science and Technology (2013K2A1A2053850).