Fifteen healthy volunteers 24.6 ± 3.2 years old (mean ± SD), were recruited for the study by online advertisements and word of mouth. Subjects that wore corrective eye-glasses were excluded from recruitment due to the glare artifact the glass generates in the wearable eye-tracking system. Subjects wearing contact lenses were not excluded from the recruitment as these lenses do not interfere with the eye-tracking system, but were required to wear them for both sessions. One participant from the sample was rejected prior to data processing due to an excessively noisy EEG signal during the reclining session which caused a loss of more than 20% stimuli for one experimental condition. Therefore the final sample was composed of 14 subjects (eight females) with an average age of 24.6 ± 3.3 years. Experimental subjects gave written informed consent prior to taking part in the study in agreement with the ethical approval obtained from the University of Liverpool Research Ethics Committee. Participants received £30 (£15 per session attended) as compensation for their travel expenses and time. All experimental procedures were conducted in two sessions and in accordance with the Declaration of Helsinki.

Face images expressing four emotional categories (fearful, disgusted, happy and neutral) were selected from the Karolinska Directed Emotional Faces (KDEF; Lundqvist et al., 1998) set and from the NimStim face inventory (Tottenham et al., 2009). The full stimulus set consisted of 180 emotional human faces, 45 images in each emotional category presented on 20 A0 poster-size sheets (841 × 1,189 mm). A total of 120 images were taken from the KDEF and 60 from the NimStim set. All 180 images used in the experiment were cropped to match in size and quality (300 DPI) using CorelDRAW software. Additionally, all images were matched for temperature and brightness and were always presented as standard portrait shots including the shoulders. A0 size panels were constructed containing nine different image stimuli presented in full color approximately 20 × 25 cm in size, with a white-on-black square fixation cross presented in the centre (14.3 × 14.3 cm) (Figure 1A). Twenty panels contained all emotional face categories pseudo-randomly distributed and maintained similar distances from the fixation cross in the centre. All panels were pasted onto Styrofoam sheets and attached to the walls using adhesive tape creating a mock gallery (Figure 1B).

EEG data was continuously recorded using a 64-channel wireless and portable EEG system (Brain Products, GmbH, Münich, Germany). Brain Products MOVE system (Brain Products, GmbH) employs a lightweight signal transmitter which participants carry on a belt (Figures 1C,D). Active shielding Ag/AgCl EEG electrodes were mounted on an electrode cap (actiCAP, Brain Products, GmbH) according to the international 10–20 electrode system, aligned to the midpoint between the anatomical landmarks of the nasion and inion, and left and right preauricular points. Electrode FCz was used as the system ground and electrodes were referenced to Cz. Electrolyte gel was applied to lower electrode-to-skin impedances to under 50 kΩ. EEG recordings were sampled at a rate of 1,000 Hz. EEG average reference was applied and the signals were digitized at 1 kHz with a BrainAmp DC amplifier linked to BrainVision Recorder program version 1.20.0601 running on Windows operating system. Data were filtered online using a 0.1–200 Hz bandpass filter.

A PUPIL (Kassner et al., 2014) binocular eye-tracking system was used to record the locations of the gaze position. PUPIL Capture software v 0.9.6 running on Ubuntu v 14.04.4 was used to generate the recordings. PUPIL eye tracker is an open source, high resolution wearable system (Figure 1C) that provides a lightweight solution for mobile gaze-tracking. Pupil locations were recorded using infra-red cameras sampling at 90 Hz with a resolution of 320 × 240 pixels. The real-world video streams were set at a sampling rate of 60 frames per second with a resolution of 600 × 800 pixels.

To calibrate the gaze locations, a manual marker 3D calibration protocol was used to generate a 9-point grid in the field of view of the participant. Calibration was repeated until gaze positions were accurate everywhere on the blank panel. Small calibration offsets occurred at times due to displacements in the wearable eye-tracker on the subject’s face. These offsets were adjusted offline using the manual gaze correction plug-in on PUPIL Player during manual tabulation of stimulus onset times. Eye-tracking data were processed using PUPIL Player v. 0.9.6 program. The ocular pupils of both eyes were located based on a centre-surround detection algorithm (Świrski et al., 2012). Exported raw gaze data is time-locked to the processing computer’s internal clock, giving millisecond precision to the eye measurements. Additionally, all recorded frames contained an accurate time stamp based on the PC processor real-time clock. Eye-tracking video files were visually inspected and stimuli onsets were manually tabulated. Each stimulus was logged on a picture-by-picture basis with stimulus onset defined as the first instance in which the gaze position landed on an image. The real times corresponding to the tabulated frames were used to import stimulus onset latencies onto the raw EEG data. Eye-tracking and EEG data streams were aligned at the start of each recording using a custom-built light-emitting trigger box, which flashed a light into the eye-tracking camera and synchronously registered a transistor logic pulse in the EEG data recording.

Volunteers attended two testing days, a standing session where stimuli were viewed while upright, and a reclining session where they sat in a recumbent position on a lounge chair. In both sessions, instructions were delivered and equipment was set up in a designated lab space following identical procedures. The order of conditions was counterbalanced across subjects with half of the participants taking part in the reclining task first, and the remaining participants taking part in the standing task first. Experimental sessions were separated by a minimum of 1 week.

Participants were fit with the EEG cap and electrode-to-skin impedances were lowered below 50 kΩ before commencing the task as well as during the break between blocks. The mobile EEG system was connected and wireless signals were visually inspected on a standing participant. Next, eye-tracking glasses were placed on the participant over the EEG cap.

A mobile base unit was assembled using a rolling trolley where the wireless signal receiver, EEG amplifier and recording computer were placed. The base unit was positioned by the experimenter, keeping a distance of no more than 7 m from the participant in order to maintain optimal signal quality. The eye-trackers were calibrated on a standing participant against a blank white panel at a distance of 1 m from the centre of the panel to match the viewing conditions during the mock gallery. Gaze-tracking was optimized using a 3D calibration routine with manual markers. The laptop registering the eye-tracking was placed in a backpack and carried by the participant for the duration of the task (Figures 1C,D). EEG cables running from the electrodes to the lightweight transmitter were also placed in the backpack to reduce cable sway artifacts (Gramann et al., 2010; Gwin et al., 2010).

Two hallways within the Eleanor Rathbone Building of the University of Liverpool were used to create a naturalistic picture gallery where the standing session took place (Figure 1B). Participants were free to view images in any order and to navigate the gallery however they chose. All subjects were asked to stand still and view each image for a minimum of a few seconds before moving onto the next image. To do so, subjects were instructed to stand at a set distance of 1 m and remain still. They first looked at the centre fixation cross before viewing each image, and were requested to return their gaze to the fixation cross before moving to the next picture. Participants only continued onto a subsequent panel after viewing all images. To maintain task engagement and to avoid drowsiness, participants were asked to select their most and least preferred face from each panel and to mark the selected pictures on a small paper-printed version of each panel. The full experiment consisted of two blocks in the mock gallery. In each block, subjects viewed 10 panels with nine images on each one. In total, participants viewed 180 different images in the experiment. On average, each of the two gallery blocks lasted approximately 15 min.

The wireless EEG was placed on the participant in the same manner as in the standing session. For the reclining task, the eye-trackers were calibrated on a blank white panel with the participant lying on the lounge chair. The blank panel was hung at a distance of approximately 1 m with 140° angle to match viewing conditions across both sessions. The inclination angle of the lounge chair was lifted to approximately 40° (Figure 1D) bringing the fixation cross in the centre to the participant’s eyeline. In this session, subjects laid on a lounge chair that provided full body support as high as the neck. A foam neck support was used to prevent electrodes on the back of the scalp smearing against the chair and provided head support. An identical 3D calibration routine using manual markers was used to optimize the gaze-tracking for the reclined condition. The eye-tracking computer was placed on the same mobile base unit where the EEG recording computer rested in this session.

Participants were instructed to view the images displayed on each panel. The experimenters hung each panel as they were completed by the participants. Subjects were instructed to relax and view each face for a couple of seconds before returning to the fixation cross and moving onto the next image. As in the standing session, no restrictions were placed on eye-movements or blinks. The same selection task was used here to maintain engagement and avoid drowsiness. Participants viewed the same 180 different images in the experiment divided into two blocks with breaks between them. Each block in the reclined session also lasted approximately 15 min.

In both sessions, electrode impedances and gaze tracking calibration were checked in the break between blocks and corrected if required. Once the viewing task was completed, the EEG cap and the eye-tracking glasses were removed. Participants were then required to rate how much they liked and if they would approach the images that had been previously seen. Ratings were performed using three visual analogue scales (VAS) sized 10 cm and anchored on each extreme. The scales evaluated the general likability of each face image as well as the level of arousal and emotional valence. Subjects were instructed to use the middle of the scale for expressions deemed neutral. (i.e., “0: Do not like” up to “100: Like very much”; “0: Unarousing” up to “100: Very Arousing” and “0: Very negative” up to “100: Very positive”). Pictures and rating scales were presented on an LCD screen using Cogent program v. 1.32 (Welcome Department of Imaging Neuroscience, United Kingdom) running on MATLAB v. R2014a (MathWorks, Inc., USA). Participants rated the same faces after each session.

EEG data were pre-processed using the Brain Electrical Source Analysis program (BESA v.6.0, MEGIS Software GmbH, Munich, Germany). Data were initially referenced using the common averaging method (Lehmann, 1987) and digital filters were applied to remove frequencies lower than 0.5 Hz and higher than 35 Hz from the EEG data to minimize electromagnetic interference (EMI) from electrical appliances. Ocular artifacts including blinks as well as vertical and horizontal eye-movement artifacts were removed using pattern matching averaged artifact topographies (Berg and Scherg, 1994; Ille et al., 2002). This method is based on a principal component analysis and creating individualized artifact topographies to maximally match and resolve eye-movement artifacts in EEG data. Additionally, EEG data was visually inspected and corrected for the presence of remaining artifacts. Trials were excluded if artifacts were present in either eye-tracking or EEG data. If participants skipped an image, failed to fixate, or gaze tracking was lost during fixation, the trial was discarded from any further analysis. To ensure consistency across our conditions, subjects with more than 20% rejected epochs for one experimental condition were eliminated from the final analysis.

Individual stimulus onset times were detected by visual inspection of the eye-tracking data. For all stimuli the first instance where a participant’s gaze made contact with a stimulus image was registered as visual onset time for event marking. Event markers were then inserted into EEG data by synchronizing the time axes of the EEG and eye-tracking system in MATLAB (The MathWorks Inc., 2018). EEG data were epoched to range from −200 ms to 600 ms relative to the first contact of the gaze with any part of a picture in each of 180 pictures. This time point effectively corresponded to part of the saccade which brought the gaze onto a particular face or object in a picture. Artifact-corrected EEG signals were then exported into EEGLAB v.14.1.1,¹ an open-source environment for processing EEG data (Delorme and Makeig, 2003). During averaging, the mean EEG activity in the baseline interval ranging from −200 ms to −100 ms was removed from each data point as it represented a more stable baseline relative to the −100 ms to 0 ms in free viewing conditions. Post-saccadic EMRPs were computed from all trials falling into eight different conditions including postural conditions (reclining and standing) and emotional facial expressions (afraid, disgusted, neutral and happy). The sampling rate of the eye-tracking system was previously calculated offline (average sampling rate: 41.1 Hz). The sampling rate was chosen based on 15-min pilot experiments to secure a continuous stream of eye-tracking data which was often discontinuous at higher sampling rates. However, due to the relatively low sampling rate, further analysis of the eye movements was not performed and eye-tracking data was used exclusively to detect stimuli onset times. EEG epochs were generated as cuts into the world video scene in a similar manner as the method described in our previous experiment (Soto et al., 2018). Artifact-corrected EEG data was then exported to EEGLAB for group-level analysis.

Individual subject data was entered into the EEGLAB STUDY structure (Delorme et al., 2011). Independent component analysis (ICA) was performed within EEGLAB for all subjects using an extended Infomax ICA algorithm to parse EEG by defining maximally statistically independent components (IC) (Dimigen, 2020; Makeig et al., 1996) using the default training parameters on EEGLAB. This method allows an accurate detection and removal of extra-cerebral sources of noise. ICA was performed on concatenated epochs for each subject individually. Event-related potentials (ERPs) and IC scalp maps were then computed across experimental conditions and used to construct a pre-clustering array to group ICs into 9 clusters. The DIPFIT2 routine from EEGLAB was used to model each IC as an equivalent current dipole by using a standardized boundary element (Oostenveld and Oostendorp, 2002). If the residual variance of the single-dipole exceeded 30%, or if the location landed outside the head, the IC was removed prior to clustering. The clustering of ICs was performed across all 14 subjects based on similarities in scalp topographies, equivalent dipole locations and ERP waveforms by means of EEGLAB’s k-means clustering routine (Lee et al., 2015). Three clusters of interest were identified and selected based on IC activity, average dipole locations and the presence of ICs in at least nine subjects. Furthermore, the two peak features in the global field power were used to identify component clusters for analysis. Cluster 2 presented 33 ICs across 13 subjects and accounted for the N170 component. Cluster 8 (16 ICs), clustered from nine subjects, presented a small peak in activity at around 90 ms and accounted for the vertex positive potential (VPP) peaking in medial scalp locations and a larger secondary peak at around 170 ms. Cluster 6 was composed of 18 ICs present in 10 subjects and represented the major contributor to the lambda component (Kazai and Yagi, 2003; Vigon et al., 2002). Statistical testing was performed on the grand averaged IC cluster activity across all experimental conditions for these three components.

The source waveforms representing the averaged clustered IC activity in each of the fitted dipoles were analysed using a 2 × 4 ANOVA for repeated measures ANOVA (Reclining vs. standing, four emotional face conditions). Specifically, a permutation-based repeated-measures ANOVA utilizing 5,000 permutations was employed (Maris and Oostenveld, 2007). Statistical testing was performed on the averaged source activity across the intervals of interest in SPSS v.22 (IBM Corp., NY, USA). We focused the analysis on 20 ms time intervals (−10 to 10 ms) relative to the peak in IC activity for each selected cluster. Post-hoc paired t-tests were performed when necessary and considered significant at p ≤ 0.05 and Bonferroni corrections for multiple comparisons were always used.

Figure 2 shows the average rating scores for each visual analogue scale across all four experimental conditions. To test the effects of body posture on the overt rating behavior, each of the three individual visual analogue scale ratings were submitted to separate 2 × 4 ANOVAs for repeated measures. Across all rating scales, we found no effects of posture observed on overt rating behavior between standing and reclining conditions [arousal scale: F_{(1, 14)} = 0.009; p = 0.927; emotion detection scale: F_{(1, 14)} = 1.112; p = 0.31; and pleasantness scale: F_{(1, 14)} = 0.251; p = 0.624]. A significant effect of emotional condition on arousal level was also found in this analysis [F_{(3, 42)} = 100.62, p < 0.001] as well as for the emotion [F_{(3, 42)} = 69.008, p < 0.001] and pleasantness scale [F_{(3, 42)} = 84.883, p < 0.001].

Pairwise t-tests were used to test differences among all pairs of emotional categories. Paired comparisons confirmed that arousal ratings were significantly different across emotional expressions specifically; disgusted expressions rated as less arousing than afraid facial expressions [t(14) = −6.1; p = 0.003] and more arousing than neutral faces [t(14) = −2.9; p = 0.012]. Unexpectedly, afraid and neutral facial expressions were rated as equally arousing by our participants [t(14) = 0.104; p = 0.92]. Similarly, the emotional scale ratings showed happy faces (80.7 ± 7.8, mean ± SD) were consistently rated as highly positive while disgusted (35.2 ± 6.6), and afraid (39.1 ± 5.7) were rated on the negative end of the scale. Neutral facial expressions, however, were perceived to be slightly negative (41.1 ± 6.8). As expected, this effect was additionally encountered in the pleasantness ratings where faces reflecting happy emotional expressions were perceived as more pleasant than disgusted [t(14) = −12.6, p < 0.001], afraid [t(14) = 14.1; p < 0.001] or neutral faces [t(14) = 10.4, p < 0.001]. Again, neutral faces were rated equally pleasant as disgusted faces [t(14) = −2.1; p > 0.05] and afraid expressions [t(14) = 0.012, p > 0.05] irrespective of the postural position.

EMRPs were constructed from −200 ms to 600 ms relative to the first instance of contact of the subject’s gaze with an image (Figure 3). Given participants remained still to view images the EEG data showed minimal neck muscle or head movement artifacts, which have been shown to affect EEG data during walking or running (Gwin et al., 2010).

IC clusters were selected on the basis of their maximum explained variance being >70% and that clusters were composed of IC from a minimum of seven subjects. Additionally, the spatial–temporal characteristics of the scalp maps and dipole locations were used to select ICs.

Three IC clusters were selected based on the peak in their averaged waveforms and scalp distributions (Figure 4). The mean IC dipole for each IC cluster of interest was located in, or near to the posterior cingulate cortex (cluster 2), the left thalamus (cluster 6) and the left angular gyrus (cluster 8). The average waveform activations of each of these IC clusters were statistically tested using a 2 × 4 (body posture × emotional condition) ANOVA for repeated measures.

Cluster 2 presents as a tightly grouped cluster of 33 ICs in occipito-temporal portions of the brain present in 13 subjects. However, we must interpret averaged cluster localizations findings with caution given that dipole localization in EEG, while informative, lacks the spatial resolution to definitively determine the precise neural generators of the observed activity.

Figure 5A illustrates the averaged IC activity across experimental conditions as well as the location of individual ICs composing clusters (Figure 4B). Due to the peak latency and estimated average location of the cluster, we have associated it to the N170 face-sensitive component. The average IC dipole location of this cluster was estimated to be located medially in the posterior cingulate cortex behind the splenium of the corpus callosum (Talairach coordinates x, y, and z: 9, −48, 23; Brodmann Area 23). The mean IC dipole accounted for 86.72% of the total explained variance (13.28% RV ± 8.8; x̄± SD). Across the different experimental conditions, cluster 2 showed peaks in mean activity occurring between 150 ms and 230 ms with an average peak at 189 ms. The ANOVA performed over the 180 ms to 200 ms time interval reflected no effects of body position on cluster 2 [F_{(1, 13)} = 0.033; p > 0.05]. No effects of emotional condition were present in the data [F_{(3, 39)} = 0.776; p > 0.05] nor was there an observable interaction effect [F_{(3, 39)} = 0.235; p > 0.05].

Cluster 8 was composed of 16 components present in nine subjects. The scalp distributions presented over the occipito-parietal regions of the scalp with a first small deflection occurring at roughly 100 ms and a main, larger peak at a latency of about 200 ms (Figure 5B). This cluster modeled the VPP which, given their co-occurrence and shared generators, has been theorized to represent the same brain mechanism as the N170 component (Joyce and Rossion, 2005). The averaged dipole explained 81.17% of the residual variance (19.83% ± 5.94; x̄ ± SD) and was located to the left angular gyrus (Brodmann Area 39, Talairach coordinates: −26, −62, 17). Statistical comparisons were performed in the time interval between 90 ms and 110 ms as well as between 190 ms and 210 ms. Within the first time interval, no significant effect of posture on evoked source activity of cluster 8 [F_{(1, 13)} = 2.603, p > 0.1] was found in the latency period surrounding the peak. Additionally, no emotional effect [F_{(3, 39)} = 0.198, p > 0.5] nor was there any effect of interactions between posture and emotional categories [F_{(3, 39)} = 0.99, p > 0.1]. A similar pattern of results are in the interval ranging from 190 ms to 210 ms with no effects of posture [F_{(1, 14)} = 0.007, p > 0.5] or emotional category [F_{(3, 39)} = 1.426, p > 0.1] nor were there any interaction effects present [F_{(3, 39)} = 1.677, p > 0.05] in the data.

Cluster 6 was comprised of 18 ICs from 10 of the subjects which were active within similar latencies as the lambda potential (Figure 6). The scalp distribution of the component shows a clear occipital distribution. The averaged IC dipole was fit to the left thalamus (Talairach coordinates −10, −6, 9) and the mean dipole accounted for 84.6% of the explained variance (16.4% ± 9.63). Cluster 6 activations showed significant effects of body posture on IC activity [F_{(1, 13)} = 6.424; p < 0.05]. The averaged IC activity was larger when subjects reclined (−0.15 ± 0.621 μV) relative to when they stood (−0.085 ± 0.415 μV). There was no effect of emotion on the mean IC cluster activity [F_{(3, 39)} = 0.475; p > 0.05] nor was an interaction effect found in this interval [F_{(3, 39)} = 0.694; p > 0.05].