Twenty-eight healthy, young (mean age 22.1 ± 2.3 years, range 19–30) subjects with no history of neurologic or neuropsychiatric disorders were recruited to participate in this study. Fourteen subjects (11 females and 3 males) participated in each task condition (emotion and shape). All subjects were right-handed according to their self-report and gave their written informed consent for participation in the study.

Subjects were tested individually in a sound-attenuated room. A computer program generated by E-Prime 2 (Schneider et al., 2002) controlled the experiment. The stimuli were presented on a 17^′′ TFT monitor (screen resolution: 1024 by 768 pixels; background color: silver – RGB: 200, 200, 200) and participants responded via the keyboard. We used three grayscale pictures (4.5 cm wide by 7.7 cm height) of human faces as prime stimuli, one for each facial expression (happy, angry, and neutral). These stimuli were taken from the NimStim Set of Facial Expressions (Tottenham et al., 2009; the reference codes of the selected faces are 20_M_HA_O, 20_M_NE_C, and 20_M_AN_O). By using photo-editing software, we created two versions of each picture, one wearing rounded glasses, and the other wearing squared glasses. As target stimuli, we used 36 Spanish words divided into two sets, one comprising 18 positive words, the other containing 18 negative words. Mean valence ratings for the words of the two sets ranged from 1.7 to 2.8 (M = 2.3) for positive words and from –0.9 to –1.8 (M = –2.3) for negative words, according to a preliminary study (N = 124; scale ranging from –3 to +3; see Sassi et al., 2014). Positive and negative words were matched for word frequency, familiarity, and word length using the LEXESP database (Sebastián-Gallés et al., 2000). Each trial consisted of the following sequence (the trial scheme is summarized in Figure 1). First, a 1000-ms fixation point (a plus sign) appeared in the center of the screen followed by one of the three prime faces, which was presented for 200 ms. Then, after an interval of 100 ms (stimulus onset asynchrony, SOA = 300 ms), a target word was shown (in capital letters and black font) and subjects indicated whether the word was positive or negative by pressing the “n” or “m” key on the computer keyboard as quickly and accurately as possible (this first response is referred to as R1). Both prime faces and target words were presented centered. The specific response-key mapping was counterbalanced across participants. Immediately following R1, a double-choice question appeared on the screen, and subjects were prompted to press, with no time limit, the key (“z” or “x”) that corresponded to the correct answer (hereafter, R2). In the emotion condition, subjects were asked whether the prime face was neutral or emotional (the emotion task), whereas in the glasses’ shape condition they were asked whether the face wore rounded or squared glasses (the shape task). The whole experiment included 72 congruent trials, 72 incongruent trials, and 144 neutral trials. In congruent trials, the prime face and the target word belonged to the same affective valence, either positive, as it happened in happy-positive trials (N = 36) or negative, as it happened in anger-negative trials (N = 36). In incongruent trials, a prime face with different valence preceded the target word, as it happened in happy-negative trials (N= 36) and anger-positive trials (N = 36). In neutral trials, a neutral prime face preceded the target word, as it happened in neutral-positive (N = 72) and neutral-negative (N = 72) trials. Target words were drawn from each set at random, with the constraint that each word appeared in two congruent trials, in two incongruent trials and in four neutral trials. A short practice block of 18 trials preceded the experimental trials.

Electroencephalography was recorded using 59 scalp channels mounted onto an elastic cap (ActiveCap, Brain Products GmbH), according to the 10–20 international system, with the reference located close to the vertex. The EEG signal was amplified (BrainAmp, Brain Products GmbH), digitized (1000 Hz sampling frequency), and filtered (0.1 to 40 Hz). The electrode impedance was kept below 5 KΩ. Four additional electrodes were placed to monitor the left/right and horizontal/vertical ocular activity. The eye movements’ artifacts were corrected with an independent component analysis (ICA) Ocular Artifact Reduction algorithm (Vision Analyzer, Brain Products GmbH). The ERPs were obtained by averaging the EEG epochs from –250 to +300 ms with respect to face onset, using the first 200 ms for baseline correction. Data were finally re-referenced using a common average reference approach.

According to previous studies, we focused on the P1 and N170 components and also on N1, which peaks in frontal regions at ∼100 ms. Additionally, our data revealed a late positive deflection, peaking at ∼240 ms, that was also investigated. Four pairs of sensors clusters, whose amplitude was calculated as the mean amplitude of their constituent sensors, were defined to model the ERP components. In each subject and for each experimental condition, the amplitudes of components’ peaks were calculated as the maximum positive/negative deflections within the time windows specified in Table 1. To better compare ERP results with source analysis results, a further cluster, conventionally not investigated in previous studies, was defined for the N170 period that covered the temporo–parietal region. These eight cluster measures were subjected to statistical analyses. In further analysis, the two occipital clusters were merged into a single cluster, and its activation was expressed in terms of the lateralization of its medial–lateral center of gravity, calculated with the following formula:

where a,b,c represent the medial–lateral coordinates of those electrodes in a 10–20 extended system.

A preliminary ICA (Hyvarinen, 1999) was performed on ERP data, which allowed for the decomposition of the signal to noise-normalized independent components (ICs). Only those ICs that showed an SNR below 1 across all intervals of interest (from –250 to 300 ms with respect to the facial onset) were removed from the ERP data (Inuggi et al., 2011a,b). The source activity was reconstructed using the cortical current density (CCD) model with a conductor volume defined by a 3-compartment boundary element method (BEM), with conductivity values of 0.33-0.0042-0.33 S/m (Fuchs et al., 2002), derived from the FSL MNI template (www.fmrib.ox.ac.uk/fsl), dimensions of 91×109×91 and a voxel size of 2×2×2 mm. The sources number (6899) and positions were obtained by sampling the cortex (5 mm wide), with their orientations fixed perpendicular to the cortical patch they originated from, and their intensities were calculated using the SWARM algorithm (Wagner et al., 2007). The CCD was reconstructed with the Curry V6 software (Neuroscan Inc., Herndon, VA, USA).

The effects of the between-subjects factor task type (emotion task, shape task) and the within-subjects factor face expression (angry, happy, or neutral) and hemisphere (left and right) over TA within each area and period were analyzed with a mixed analysis of variance (ANOVA). The Kolmogorov–Smirnov test was used to examine the normal distribution of the data, and, when appropriate, the Greenhouse–Geisser correction was applied. The significance level of the main effects (task type, emotional expression, and hemisphere) and their interactions were corrected for multiple comparisons (14 ROI × 3 periods) using a false discovery rate (FDR) approach, but using a more conservative version (Benjamini and Yekutieli, 2001) compared to standard FDR. According to its formula (α/Σ_i=1..k(1/i), where i = 42 is the number of multiple comparisons and α = 0.05 is the predetermined p-value), we report only the significant p-values below 0.0112. Because the number of multiple comparisons was lower in the ERP analysis (8 cluster × 3 periods), the corrected threshold was 0.0132. The size effects were reported through the ηp2 value. Post hoc comparisons of within-subjects (facial expression) and between-subjects (task type) factors were performed with paired and unpaired t-tests. The multiple pairwise comparisons of facial expressions were adjusted with the Bonferroni correction.

To provide the ERP equivalent of our source analysis results, a mixed ANOVA, analyzing the effects of task type and face expression, was also performed over the ERP electrode clusters that overlay the ROI of the sources significantly affected by our experimental factors.

Trials with incorrect responses to the target word (R1; 1.8 and 1.5% for the emotion task and the shape task, respectively), and trials with incorrect responses to the to-be-attended facial feature (R2; 3.1 and 5.0% for the emotion task and the shape task, respectively) were excluded from analysis. In addition, we excluded trials with RTs below 200 ms (anticipations) or more than three standard deviation (omissions) from the subject’s mean for each condition (1,90%). The mean RT for R1 in the emotion task was 790 ms (SD = 144) for congruent trials (happy face/positive word and angry face/negative word trials) and 825 ms (SD = 164) for incongruent trials (angry face/positive word and happy face/negative word trials). In the shape task, the mean RT was 767 ms (SD = 173) for congruent trials and 768 ms (SD = 155) for incongruent trials. These means were submitted to mixed ANOVA with congruency (congruent, incongruent) and task type (emotion, shape) as factors. There was a main effect of congruency, F(1,26) = 9.75; MSE = 464; p = 0.004; ηp2 = 0.27, revealing that responses were faster for congruent than for incongruent trials (this difference represents the affective priming effect, M = 18 ms). However, this effect was qualified by a congruency by task type interaction, F(1,26) = 9.10, MSE = 464, p = 0.006, ηp2= 0.26. Post hoc Fisher’s least significant difference (LSD) tests (MSE = 25411, df = 26,479) revealed significant congruency effect for the emotion task (priming effect = 35 ms, p < 0.001) but no effect at all for the shape task (priming effect = 0.6 ms, p = 0.941). Results, thus, replicate those obtained in our previous behavioral study (Sassi et al., 2014). To support the goodness of our protocol, we verified that neither the main effects of word valence, F(1,26) = 1.56, p = 0.222, and task type, F < 1, nor their interaction, F(1,26) = 1.30, p = 0.26, were statistically significant with the neutral expression. Analysis of error rate (CR1) revealed no statistically significant effects.

The group averages of the evoked potentials elicited by the two tasks, which merged the three emotional faces, are displayed in Figure 3. Table 2 summarizes the activation’s center of gravity coordinates and PL values of the ROIs, where a significant effect of either task type or face emotion could be observed.

During the P100 component, the medial–lateral center of gravity of the cluster obtained by merging the right and left occipital clusters was modulated by task type [F(1,26) = 5.45, p= 0.011, ηp2 = 0.25], which was more right-lateralized in the emotion task (M = 11.3, SD = 4.5 mm), than in the shape task (M= –0.9, SD = 3.8 mm). At ∼170 ms, the occipito-temporal cluster that overlays the lateral BA37 was not affected by the task type. In the right occipito-temporal cluster, which should provide the ERP equivalent of the right TPJ activation, a significant interaction was found between task type and facial expression in the occipito-temporal cluster [F(1.52,21.13) = 5.20, p= 0.012, ηp2 = 0.24]. Nevertheless, we found a trend (p= 0.065) versus a more negative peak to angry faces compared to neutral ones in the shape task (Figure 7). No differences emerged within the parieto-temporal cluster.

In the present study, we confirm that the ventral stream is highly modulated by the observer’s attention. The activity of the occipital areas at ∼100 ms was more right lateralized in the emotion task than in the shape task and, more notably, when subjects attended to the facial expression, activation produced by emotional face expressions was more right lateralized than activation produced by neutral faces. Such selectivity disappeared when subjects attended to the glasses’ shape.

Considering that the assessment of FFA activity through scalp recordings is widely questioned, as the area lies within the inferior part of the temporal cortex, we created the lateral BA37 ROI because previous neuroimaging studies showed a correlation between the N170 EEG component, calculated by electrodes overlaying it and fMRI-derived FFA activity (Horovitz et al., 2004; Sadeh et al., 2010), which suggests that surface electrodes may capture at least part of FFA activity. Additionally, electro-corticography studies have revealed that lateral BA37 is also involved in face processing (Rossion et al., 2003; Tsuchiya et al., 2008). At ∼170 ms, lateral BA37 activation was reduced in the shape task compared with the emotional task, which suggests that when subjects were asked to ignore the facial expression and just concentrate on the glasses’ shape, the detailed face features might not have been very distinctive. This result agrees with previous findings that report larger activity in FFA for faces compared to non-face objects (Haxby et al., 2000; Rossion et al., 2003). Taken together, our behavioral and neurophysiological results strongly suggest that our shape task succeeded in guiding subjects’ attention away from any face feature, preventing any conscious monitoring of the emotional content of the face. In the long debate over the pre-attentive automaticity of emotional processing, our results suggest that an appropriate level of attention is needed to process emotional expressions. Although presented in the same visual focus, the reduced BA37 activity and the loss of emotional selectivity of primary visual areas in the shape task suggest that subjects presumably focused their attention just on the glasses’ shape and ignored the underlying emotional expression.

The lateralization of the activations found in the present study deserves further comments. The lateralization of emotional processing is still an open issue because the two main theories, supporting either the right-hemisphere hypothesis (RHH; Borod et al., 1998; Bourne, 2010) or the valence-specific hypothesis (VSH; Mandal et al., 1991; Adolphs et al., 2001), have been questioned by more recent fMRI meta-analysis investigations (Fusar-Poli et al., 2009; Sabatinelli et al., 2011). The bulk of evidence shows bilateral activation for emotional face processing in most emotion-related areas, although lateralization might be modulated by gender (see for example Wager et al., 2003). In the present study most (22 out of 28) of the subjects were women, and our data are consistent with a previous EEG report specifically investigating the gender effect over emotional face processing. Proverbio et al. (2006) in fact found maximal P1 amplitude over the right occipital cortex in both genders, consistently with our results showing that in the emotion task occipital activity around 100 ms was right lateralized. The lack of a right lateralization observed in our data during the N170 may appear inconsistent with the widely accepted right predominance of FFA in face processing (Kanwisher and Yovel, 2006). However, this again agrees with Proverbio et al.’s (2006) findings of a right lateralization of N170 only in men. In contrast, women exhibited a bilateral pattern. These results can help foster better understanding of the inconsistencies in the literature on the right hemisphere advantage in the occipito-temporal cortices when processing faces and confirm the relevance of incorporating gender information.

Although both static and emotional features appeared under-processed by canonical face processing cortical areas, unattended angry expressions were able to activate the TPJ, a cortical expanse implicated in a wide spectrum of high-order cognitive functions ranging from social cognition (Saxe and Kanwisher, 2003) to attention selection (Corbetta and Shulman, 2002). The latter branch of investigation showed that the TPJ is part of the VAN, a fronto-parietal network that, during focused activities, is formally involved in re-orienting (shifting) attention to stimuli relevant to the immediate goal. Nevertheless, because the attentional focus covered a similar area in both tasks, no reorienting process was expected, as our IPS activity also indicates. The latter is in fact part of the DAN, which contains the proper circuitry to implement the focus reorienting, and was not modulated by our experimental conditions. The absence of any modulation over frontal areas might be interpreted accordingly; the integration between ventral and DANs, needed for attention re-orienting, occurs in such aforementioned frontal areas where the two networks highly overlap (Fox et al., 2006).

Thus, the present findings support the proposal that VAN activation, at least in its parietal areas, might not exclusively be involved in attentional reorienting. It is consistent with more recent reports that suggest that TPJ activity might be triggered by both external sensory stimuli and internal memory-based information, thus providing bottom–up signals to other systems about relevant stimuli for further inspection (Cabeza et al., 2012). In agreement with the present results, VAN activity has also been observed when behaviorally relevant, rather than salient, stimuli are presented while the individual is engaged in another task (Corbetta et al., 2008). Accordingly, the activation of TPJ just when the unattended face was shown with an angry expression suggests that negative emotions can pre-attentively evoke bottom–up cortical signals, according to their behavioral relevance, even when attention is focused on emotion-irrelevant features in a task that we assumed exhausted the attentional resources to process the emotional content of faces. Because the ventral stream and STS were not modulated by the degree of unattended emotional content and the VAN is considered a supramodal network (Macaluso et al., 2002; Green et al., 2011) not able to decode the threatening pattern from facial expression, we suggest that TPJ activation might be triggered from other brain regions. Several neuroimaging studies suggested that, in parallel with the cortical stream (Palermo and Rhodes, 2007), a subcortical pathway, that reaches the amygdala through fast and coarse subcortical inputs that originate in the superior colliculus and finally project onto fronto-parietal areas, is thought to implement a brain circuitry specialized in emotional attention (Vuilleumier, 2005). This circuitry, likely partly modulated by the attentional focus (Pourtois et al., 2013) is involved in the rapid and automatic detection of negative facial expressions (for review; see Vuilleumier and Pourtois, 2007), and it seems to play a crucial role in directing attention and information processing to threatening stimuli (Ohman and Mineka, 2001). Because reconstructing amygdala activity with EEG presents several accuracy limitations, as it will be discussed later, further studies that integrate EEG with neuroimaging techniques are surely needed, but our data are consistent with such a model. A previous MEG study showed in fact that the amygdala activates as early as 100 ms after stimulus presentation (Streit et al., 2003), a latency early enough to trigger TPJ activation at ∼150–170 ms. The present TPJ activation of ∼170 ms is consistent with a recent ERP study that investigates the threat detection advantage (Feldmann-Wüstefeld et al., 2011), which revealed that angry and happy expression processing started to differ at ∼160 ms. This suggests that angry faces may trigger a fear module that enables their rapid processing and recruit additional attentional resources, possibly by means of TPJ, as is here hypothesized.

In conclusion, within the VAN, TPJ activation at this early latency primarily signals the behavioral relevance of a task-irrelevant aversive stimulus, irrespective of whether that stimulus requires a physical shift of attention (involving the dorsal network). The fact that such a trigger was not followed by an actual over-processing of face features is likely due to the task demands that, immediately after face offset (∼200 ms), required that subjects focus on word onset and the corresponding response related to its emotional valence.

In the present paper, we aimed to provide an ERP-equivalent of the activations produced by source analysis. We thus focused this analysis only on the time windows and clusters that surround the cortical areas affected by our experimental conditions. ERP analysis found that attended emotions, compared to ignored emotions, have their occipital P1 peak more right lateralized but was unable to assess the selectivity toward attended emotional faces, which disappeared in the shape task. In a similar manner, ERP analysis could detect the interaction between task and emotions at ∼170 ms in the right occipito-temporal cluster, but it did not find a significant difference between angry and non-angry ignored faces. Of course, the current ERP approach is only one of many possible approaches. We are not concluding that another ERP analysis would have been unable to locate the same effects found with source analysis. However, even if such an effect had been encountered in a cluster or in a channel (e.g., CP4 or CP6), it would have been impossible to clearly attribute it to one of the areas beneath and close to the sensors cluster. Ideally, both pSTS and BA37 would have been valid candidates, and we could have argued that because they are part of the cortical stream supposedly deputed to extract face features, they would have presumably shown such functioning also in the attended condition, but that doubt would have persisted, and the involvement of TPJ could have been just one of the possible hypotheses. Instead, source analysis, when calculating the center of gravity of the large ROI covering the temporal and parietal lobe, indicated the TPJ involvement.

The main limits of EEG source analysis are its high sensitivity to artifacts, the low signal-to-noise ratio and the limited spatial resolution. To properly address these limits, we employed a consolidated methodological approach (Inuggi et al., 2010, 2011a,b; Gonzalez-Rosa et al., 2013), which has consistently proved to obtain results in line with the neuroimaging literature. We used a seed-based analysis instead of a voxel-wise one because this approach is often used in both EEG and neuroimaging analyses, when strong hypothesis of the involved brain areas is possible. In fact, although the experimental task, seen as a whole, is brand new, the areas involved in the investigated interval have been accurately described in the past as producing a consistent picture that guided and supported our ROI selection. We adopted a conservative approach, selecting ROIs in areas on the outer surface of the brain where the spatial resolution of the EEG source analysis is maximal and avoiding the investigation of deep brain areas such as the proper FFA, orbitofrontal, para-hippocampal cortices and amygdala. These areas were reported in several neuroimaging studies but their reconstruction through EEG presents several methodological issues. EEG source analysis accuracy is in fact highly corrupted by the huge anisotropy and inhomogeneity of the brain that blur the emerging signal when it is not modeled by a proper volume conductor model. Deep sources are of course more buried within the brain as the ideal lines separating the sources from the scalp electrodes cross much more tissues of different conductivities than superficial sources, making the blurring much higher. Concerning the temporal selection, we opted to analyze up to ∼300 ms because we were interested in assessing the automatic processing of face stimuli, aware of the fact that the later components would have been altered by subjects’ intentions or strategies to concentrate on target stimulus (the word) decoding. We decided to investigate the task effect as a between-subject factor as we were interested in maximizing the “unattendeness” of face emotional expressions in the glasses shape task as much as possible. We feared that if half of subjects, due to the counterbalance of the task order, performed the emotional task first and then the glasses task, facial emotion might have acquired some relevance even when the glasses task asked subjects to attend and respond to only the glasses’ shape. In addition, we would have obtained an incredibly long task, with unpredictable consequences over subjects’ attention and performance level, with the risk of introducing undesired biases into our results. The failure to locate the areas that actually discriminate and extract the emotional features of the faces surely represents a limit of the present exploration. A trend versus a higher activation of pSTS in emotional compared to neutral expressions was found only in the emotion task. However, it was not significant even before applying the Benjamin and Yekuteli correction. This might be due to the spatio-temporal resolution of the method here implemented or, more presumably, because emotional processing also involves deep brain areas, such as FFA s, orbitofrontal cortices and subcortical regions.