Twenty-eight right-handed healthy female subjects participated in the study after exclusion of acute or past neurological, psychiatric, or endocrine illness and use of psychotropic/endocrine medication. This study was carried out in accordance with the recommendations of Hannover Medical School's Ethics Committee with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the internal Ethics Committee of Hannover Medical School. We analyzed functional imaging data, eye tracking data and behavioral data from 18 participants (mean age (±SD) 24.8 ± 3.5 years) after excluding 10 participants due to either technical problems during functional imaging (n = 2), above cut-off depression scores (>14 in the Beck Depression Inventory (BDI); n = 1), low overall eye tracking rates (<80%; n = 6) or non-compliance with the instruction to fixate the center of the screen (gaze shifts on 47.5% of hit trials; n = 1).

Participants attended one session of functional imaging with concomitant eye tracking. Before and after the scanning session, participants filled in questionnaires to assess changes in state anxiety (STAI-S) and positive and negative affect (PANAS) related to the experimental procedure. Furthermore, participants completed versions of the BDI, the State Trait Anxiety Inventory (STAI-T) and the Toronto Alexithymia Scale (TAS-20) to rule out effects of altered mood states or emotional perception.

To assess the processing of unattended emotional facial expressions in the left, central, or right visual field, we used an event-related design presented in Figure 1. Photographs of fearful, happy and neutral faces taken at a straight angle were taken from the Karolinska Directed Emotional Faces database (KDEF; Lundqvist et al., 1998), transformed into grayscale and clipped by an oval shape to reveal the face. Within each of the nine experimental conditions (3 valences × 3 locations), 22 stimuli were presented containing equal proportions of male and female faces. In total 231 trials were shown, including 33 null events. Stimulus presentation was randomized and different for each participant. Each stimulus appeared for 1 sec followed by an inter-stimulus interval of 2,500 ± 750 ms. A white crosshairs was shown at the center of the screen throughout the entire experiment. Participants were instructed to look only at the fixation cross and not let their gaze be diverted by any of the stimuli. To ensure a continuously high level of attention, participants indicated the position of each stimulus by pressing a response button corresponding to the left (left index), center (right thumb), or right (right index) of the screen. Before functional imaging, participants performed a training session outside of the scanner where they viewed a different set of emotional faces.

By tracking movements of the participants' right eye, we validated compliance with the fixation instruction during scanning. For this purpose, a built-in eye tracker attached to the right ocular of the binocular video goggles (both Nordic NeuroLab, Bergen, Norway) continuously tracked the gaze direction of the participants at a rate of 60 Hz. Button presses were recorded using MR-compatible response grips (Nordic NeuroLab). For stimulus presentation and feedback recording, we used Presentation® (Neurobehavioral Systems, Inc.). Eye movement data were recorded using iView X MRI-SV (SMI, Potsdam, Germany).

For functional image acquisition we used a 1.5 T Siemens Avanto (Siemens, Erlangen, Germany) equipped with a 12-channel array head coil. By means of a T2*-weighted gradient-echo echo-planar imaging sequence, we acquired 26 axial 3 mm slices with a gap of 40% resulting in an in-plane resolution of 3 × 3 × 4.2 mm3 (TR/TE = 1530/30 ms, FA = 90°, FOV = 192 mm, 64 × 64 matrix). We employed the in-built Siemens iPat mode (integrated parallel acquisition technique) and accelerated the sequence by a factor of 2 using GRAPPA [generalized autocalibrating partially parallel acquisition (Griswold et al., 2002)]. The slices were tilted by about −45° along the line from the lower boundary of the orbitofrontal cortex (BA 11) to the lower boundary of the cerebellar nodule to ameliorate susceptibility artifacts and limit partial volume effects due to tissue borders in the amygdala and temporal cortex (Merboldt et al., 2001; Weiskopf et al., 2006). Subsequent visual inspection of all data sets confirmed good coverage of the amygdala and temporal regions. We collected additional high-resolution anatomical images from each participant using a T1-weighted MPRAGE sequence (176 sagittal slices, 1 mm slice thickness, GRAPPA acceleration factor 2).

Reaction times of hit trials (ms) and error rates (percent) were calculated for each participant and each experimental condition and compared in a repeated-measures ANOVA with factors face valence (fearful, happy, neutral) and face location (left, central, right) using SPSS 17.0 and a significance level of p < 0.05 two-sided. We calculated post-hoc tests for significant main effects and interactions at a one-sided threshold of p < 0.025.

We assessed task compliance during hit trials using BeGaze 3.0 (SMI) by analyzing means and standard deviations for the frequency of gaze shifts and location of fixations away from the central crosshairs along the horizontal axis (i.e., to the left or right). Frequencies of gaze deviations during hit trials were then binned into four categories (Figure 2A) corresponding to fixations falling into the space between (1) the fixation cross and the center of the inner eye of centrally presented faces (bin 1: >20 to 40 pixels), (2) the center of the inner eye and the inner border of peripherally presented faces (bin 2: >40–80 pixels), (3) the inner border and the center of peripherally presented faces (bin 3: >80 to 200 pixels) and (4) beyond the center of peripherally presented faces (bin 4: >200 pixels). The percentage of hit trials during which the eye tracker registered gaze deviations along the horizontal axis were compared across experimental conditions by means of a repeated-measures ANOVA with the factors gaze direction (left, right), stimulus valence (fearful, happy, neutral), stimulus location (left, central, right), and gaze distance (bin1, bin2, bin3, bin4) using SPSS 17.0 at a two-sided threshold of p < 0.05.

Before functional data analysis in SPM 8 (http://www.fil.ion.ucl.ac.uk/spm) we discarded the first three volumes of each data set to allow T1 saturation to reach a stable level. Data were realigned to the eighth slice of each volume containing the center of the amygdala as our primary region of interest and then slice timed to reduce variance attributable to motion and differences in slice acquisition time. We coregistered the high-resolution anatomical image to the functional mean image and segmented the coregistered structural image using the gray and white matter tissue probability maps provided in SPM8. Subsequently, we normalized the functional data at a voxel size of 3 × 3 × 3 mm using the normalization parameters from the segmentation procedure. Spatial smoothing with an isotropic Gaussian kernel of 9 mm FWHM reduced high-frequency noise. For each participant, t-statistics for the nine experimental conditions [fearful left (FL), fearful central (FC), fearful right (FR), happy left (HL), happy central (HL), happy right (HR), neutral left (NL), neutral central (NC), neutral right (NR)] were calculated in a first-level fixed effects model corrected for serial autocorrelations (AR1) and low frequency signals (high-pass filter of 128 s). To this end, we modeled boxcar regressors for hit trials from the experimental conditions according to the actual length of the stimuli and convolved these with the canonical hemodynamic response function implemented in SPM8. Regressors of no interest included error trials, as well as the individual realignment parameters. A random effects group level analysis compared interactions of the experimental factors face valence (fearful, happy, neutral) and face location (left, central, right) in a full-factorial within-subject GLM at an intensity threshold of p < 0.001 (uncorrected) with an extent cluster threshold of k = 4 as calculated by SPM8. Peak coordinates of activated clusters are reported in Table 2 in MNI space along with their corresponding anatomical labels according to the AAL database within the WFU pickatlas 3.0.3 (Tzourio-Mazoyer et al., 2002; Maldjian et al., 2003, 2004).

We extracted contrast estimates for each experimental condition from clusters showing sensitivity to both valence and location and performed post-hoc tests on these values for further exploration of the neural activation patterns. In SPSS 17.0 we compared experimental conditions by means of one-sided t-Tests for dependent samples at an uncorrected p < 0.025 and at p < 0.0028 Bonferroni-corrected.

The average error rate was 1.49% (SD = 1.32). A 3 × 3 ANOVA with factors face valence (afraid, happy, neutral) and face location (left, center, right) yielded a main effect of location [Figure 2B; F_{(2, 34)} = 13.4, p < 0.001, ηp2=0.440]. Post-hoc comparisons of mean RT with respect to location show faster responses to stimuli presented in the left hemifield both compared with stimuli presented centrally [t₍₅₃₎ = 7.36, p < 0.001] and in the right hemifield [t₍₅₃₎ = 4.74, p < 0.001]. Moreover, RT to stimuli presented in the right hemifield were faster compared with RT to centrally presented stimuli [t₍₅₃₎ = 2.49, p = 0.016].

The main effect of face valence was non-significant [F_{(2, 34)} = 0.450, p = 0.641, ηp2=0.026]. The main effect of face location was not qualified by an interaction between face valence and face location [F_{(4, 68)} = 2.22, p = 0.076, ηp2=0.115]. Means and standard deviations for error rates and RT for each of the nine experimental conditions are reported in Table 1. Means and standard deviations for the analysis of RT are also given in the Supplementary Material.

The average tracking rate of hit trials was 97.4% (SD = 5.3). The average percentage of trials with gaze deviations away from the central crosshairs was 10.8% (SD = 11.8). Eye movement data were also analyzed using a within-subjects ANOVA with factors gaze direction (left, right), face location (left, center, right), face valence (afraid, happy, neutral) and gaze distance (bin1, bin2, bin3, bin4). Degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity when Mauchly's test indicated that the assumption of sphericity had been violated. Asterisks mark the respective contrasts.

This analysis revealed a main effect of face location [F_{(2, 34)} = 4.81, p = 0.014, ηp2=0.221]. More gaze shifts occurred during trials with peripherally presented stimuli compared with centrally presented stimuli (left vs. center p = 0.015; right vs. center p = 0.037).

We also found a main effect of gaze distance [F_{(3, 51)} = 3.95, p = 0.013, ηp2=0.168]. More gaze shifts fell into the first three bins compared with the most distal bin (bin 1 vs. 4 p = 0.030, bin 2 vs. 3 p = 0.010, bin 3 vs. 4 p = 0.002).

These effects were qualified by two-way interactions of face location × gaze distance [F_{(2.7, 45.2)} = 5.01, p = 0.006^*, ηp2=0.228] and gaze direction × face location [F_{(1.4, 23.8)} = 9.82, p = 0.002^*, ηp2=0.366], as well as by a three-way interaction of gaze direction × face location × gaze distance [F_{(2.4, 39.9)} = 7.29, p = 0.001^*, ηp2=0.300].

There was no significant main effect or interaction with stimulus valence (all F ≤ 1.78, p ≥ 0.150, ηp2≤0.093) or any other main effects and interactions (all F ≤ 0.39, p ≥ 0.668, ηp2 ≤ 0.023). All means and standard deviations for gaze deviations along the horizontal axis are reported in Table 1. Means and standard deviations for the analysis of eye movements are also given in the Supplementary Material.

Focussing on the interaction of face location × face valence on the whole brain level, we found activation in the left temporal pole, left middle temporal gyrus, right amygdala, right posterior putamen and the right superior frontal gyrus (uncorrected p < 0.001, k = 4). In a post-hoc analysis we compared mean contrast estimates from the activated clusters using one-sided t-Tests with a statistical threshold of p < 0.025 (comparisons that are also significant at a Bonferroni-corrected p < 0.0028 are marked by asterisks; see Figure 3). In this analysis, we found a subset of the above-mentioned regions to be sensitive to the central presentation of fearful faces. This subset includes the left temporal pole [Figure 3A; FC>FL: t₍₁₇₎ = 4.91, p < 0.0001^*], the right amygdala [Figure 3B; FC>FR: t₍₁₇₎ = 3.36, p = 0.0019^*; FC>FL: t₍₁₇₎ = 2.75, p = 0.0069] and the left middle temporal gyrus [Figure 3C; FC>FL: t₍₁₇₎ = 2.70, p = 0.0075]. The following regions responded to neutral faces presented in the periphery, especially in the left hemifield: the right amygdala [NL>NC: t₍₁₇₎ = 3.08, p = 0.0034; NR>NC: t₍₁₇₎ = 2.12, p = 0.024], the left middle temporal gyrus [NL>NR: t₍₁₉₎ = 5.01, p < 0.0001^*], the right posterior putamen [Figure 3D; NL>NC: t₍₁₇₎ = 3.88, p = 0.0006^*; NL>NR: t₍₁₇₎ = 2.34, p = 0.0159], and the right superior frontal gyrus [Supplementary Figure 1; NL>NC: t₍₁₇₎ = 3.02, p = 0.0039; NL>NR: t₍₁₇₎ = 2.11, p = 0.025]. Happy faces presented in the left hemifield activated the right posterior putamen [HL>HR: t₍₁₇₎ = 2.27, p = 0.0183] and the left temporal pole [HL>HC: t₍₁₇₎ = 4.67, p = 0.0001^*; HL>HR: t₍₁₇₎ = 3.86, p = 0.0006^*]. Main effects of face location and face valence are shown in the Supplementary Material.

In the present study, the right amygdala was sensitive to centrally presented fearful faces and to neutral faces presented in the left hemifield. The amygdala has a well-documented role in the processing of emotional and neutral facial expressions (Haxby et al., 2002; Platek et al., 2006; Olson et al., 2007) even when they are unconsciously perceived (Vuilleumier, 2005). The ability to rapidly detect potentially relevant stimuli signaling the need for withdrawal are, at least in part, related to a subcortical circuit comprising of the amygdala, the superior colliculus and the pulvinar nucleus of the thalamus (Morris et al., 1999; LeDoux, 2000, 2007; Tamietto and de Gelder, 2010). Behavioral studies of unconscious emotional processing using masked stimuli also revealed enhanced autonomic responses to fearful faces presented in the left visual field compared with fearful faces presented in the right visual field or neutral faces presented in either field (Kimura et al., 2004).

Several authors found different neural correlates for explicit vs. implicit threat processing (Critchley et al., 2000; Habel et al., 2007; Tamietto and de Gelder, 2010; Almeida et al., 2013). While the brainstem and temporal regions such as the superior temporal gyrus were especially responsive during explicit tasks (Critchley et al., 2000), the amygdala was involved in both explicit (Habel et al., 2007) and implicit tasks (Critchley et al., 2000).

Complex perceptual inputs and visceral responses are integrated e.g., in the temporal pole (Olson et al., 2007). The present results corroborate the proposed involvement of the amygdala/temporal pole complex in the perception of (potential) threat signals, such as centrally perceived fearful faces and peripherally perceived neutral faces.

Studies directly comparing central and peripheral vision point to a central bias for both human and animal faces (Liu and Ioannides, 2010; Morawetz et al., 2010; Almeida et al., 2013). As defined by Strasburger et al. (2011) central/foveal vision refers to the perception of objects within 2° of eccentricity, while the perception of objects presented beyond 2° of eccentricity is classified as peripheral vision. In a functional imaging study Almeida et al. (2013) systematically manipulated spatial location (central or peripheral) and stimulus type (face, real or fake snake shape) as a main outcome measure while using both implicit and explicit threat tasks. This study found foveal representations in subcortical structures such as the amygdala, the pulvinar nucleus of the thalamus and the superior colliculus. As these regions played distinct roles in the central and peripheral processing of snake shapes the authors concluded that there are multiple phylogenetic fingerprints in the subcortical responses to fear-relevant stimuli. In the present study, a distributed network including the amygdala, posterior putamen, temporal cortex and superior frontal cortex was sensitive to face valence of stimuli presented at foveal, and peripheral locations. These findings are in line with the view that stimuli such as faces demand detailed processing whereas peripheral object processing is based on a rather coarse identification (Kanwisher, 2001; Levy et al., 2001). In this context, the amygdala receives direct input from ventral areas that show a known bias toward foveal (central) input.

Behavioral studies suggest that parafoveal and peripheral locations entail relatively sound performance in stimulus recognition and categorization (Malach et al., 2002; Calvo and Lang, 2005). This is especially true for faces which can be efficiently processed and detected when presented outside of the fovea (Carretié, 2014). Investigating how the amount of stimulus eccentricity influences behavioral and neural responses, Rigoulot et al. (2011) created an experimental setup comparing the categorization of fearful and neutral faces at four peripheral locations (15° and 30° to the right and left) in a female sample. Less eccentricity as well as more negative emotional valence detrimentally influenced performance in that reaction times were shorter for fearful faces and for faces presented at 15° eccentricity. Fearful faces also entailed greater N170 amplitudes, even at far eccentric locations, further corroborating the notion that the preferential neural coding of fearful expressions persists even in far peripheral vision.

The behavioral substrate of a visual field bias in the present study is in line with split-field and lesion studies reporting a left visual field superiority for the perception of emotional facial expressions in general (e.g., Mandal and Singh, 1990; Mandal et al., 1991). Our results corroborate the view that the evaluation of emotional facial expressions happens automatically even when this feature is task-irrelevant (Pessoa, 2008). This automatic evaluation does not necessarily seem to entail a detrimental effect on task performance as suggested by the lack of valence-specific effects on both response times and eye movements in the present study.

In addition to the neural responses to threat-related stimuli, we also found specific responses to neutral and/or approach-related stimuli. When faces appeared in the left hemifield, the left temporal pole and the right posterior putamen responded to happy faces; the left middle temporal gyrus, the right amygdala, the right posterior putamen and the right superior frontal gyrus responded to neutral faces. These results complement previous research finding that salient face stimuli activate the middle temporal gyrus (Haxby et al., 2002; Platek et al., 2006), putamen (Leveroni et al., 2000; Fu et al., 2007) and superior frontal gyrus (Ebner et al., 2012). Our results also corroborate previous evidence regarding the endogenous (top-down) attention capturing quality of emotional stimulation when subjects perceive emotional stimulation outside their attention focus (Carretié, 2014).

Our functional imaging results yielded an activation pattern that is different from the “valence hypothesis” stating that the right hemisphere is specialized in processing negative emotions whereas the left hemisphere is specialized in processing positive emotions (Davidson, 1992, 1995). In humans, lateralization seems to be quite strict at the primary sensory level (e.g., contralateral processing of visual input in the primary visual cortex) but less obvious when it comes to more complex cognitive functions such as face processing (see Phan et al., 2002; Murphy et al., 2003; Wager et al., 2003 for review). Sung et al. (2011) suggest that a static circuit similar to passive information filtering is responsible for the involvement of low-level visual areas. Depending on the task, higher areas might dynamically modulate these neural circuits, as a right hemispheric (i.e., left visual field) superiority for facial processing differs with respect to stimulus content such as the orientation of faces in the picture (upright or inverted). Based on others' as well as our own previous findings (Preibisch et al., 2009) we suggest that functional hemispheric specialization in face processing does not necessarily require lateralization to one hemisphere. Rather, preferred pathways for emotional contents are required which may depend, among other factors, on the side of input and the emotional content of the stimuli. This claim is supported by the fact that we found the network that processes potential threat signals to be sensitive to both valence category and stimulus location. A set of regions including the right amygdala, left temporal cortex, right posterior putamen and right superior prefrontal cortex responded to centrally presented fearful faces and/or peripherally presented happy and neutral faces especially in the left visual hemifield suggesting a neural substrate of a visual field bias (Asthana and Mandal, 2001; Vauclair and Donnot, 2005; Proverbio et al., 2006).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.