It is well established that human faces induce stronger involuntary orienting responses than other visual objects. Yet, the timing of this preferential orienting response at the neural level is still unknown. Here, we used an antisaccade paradigm to investigate the neural dynamics preceding the onset of reflexive and voluntary saccades elicited by human faces and nonface visual objects, normalized for their global low-level visual properties. High-density event-related potentials (ERPs) were recorded in observers as they performed interleaved pro- and antisaccades toward a lateralized target. For reflexive saccades, we report an ERP modulation specific to faces as early as 40–60 ms following stimulus onset over parieto-occipital sites, further predicting the speed of saccade execution. This was not linked to differences in the programming of the saccadic eye movements, as it occurred early in time. For the first time, we present electrophysiological evidence of early target selection to faces in reflexive orienting responses over parieto-occipital cortex that facilitates the triggering of saccades toward faces. We argue for a 2-stage process in the representation of a face in involuntary spatial orienting with an initial, rapid implicit processing of the visual properties of a face, followed by subsequent stimulus categorization depicted by the N170 component.
Our ability to respond automatically to external events and to voluntary suppress responses to irrelevant cues requires flexible behavior. This flexibility can be assessed easily within the visual saccadic system using pro- and antisaccade paradigms as they elicit different behavior patterns in response to the same stimulus. The prosaccade task requires participants to look toward a suddenly appearing stimulus in the periphery, whereas the antisaccade task requires participants to look away from the stimulus by generating an eye movement in the opposite direction to its mirror location. To correctly perform an antisaccade, the neural process that triggers the reflexive, stimulus-driven response to the target must be inhibited, so that a voluntary, task-driven saccade can be generated in the opposite direction (Hallet 1978; Connolly et al. 2000; Munoz and Everling 2004). Antisaccade tasks thus represent very sensitive means of dissociating voluntary from reflexive (automatic) orienting responses. Regarding the neural correlates of antisaccades performance, event-related functional magnetic resonance imaging (fMRI) and human lesion studies have revealed that the dorso-lateral prefrontal cortex (DLPFC) and supramarginal gyrus (SMG) are likely to be involved in the reflexive saccadic inhibition, whereas the frontal eye fields (FEFs) and intraparietal sulcus (IPS) are more selectively involved in the preparation and generation of voluntary antisaccades (Pierrot-Deseilligny et al. 2002; Connolly et al. 2002; Cornelissen et al. 2002; DeSouza et al. 2003; Ploner et al. 2005; Brown et al. 2006; Ettinger et al. 2008).
While the antisaccade task has been studied intensively with abstract stimuli (mainly dots) that can be viewed as being somewhat remote from real word object recognition, to date, very few studies have employed this paradigm with more realistic high-level stimuli (Gilchrist and Proske 2006; Morand et al. 2010). Human faces, among other visual objects, are arguably the most salient visual stimuli we process every day, conveying crucial information for social interactions, and are thus likely to have a special impact on saccadic orienting responses. Indeed, we recently reported a significant increase in antisaccade error rates for faces as well as faster prosaccades to faces compared with other visual categories, indicating that human faces induced stronger involuntary orienting responses than other visual objects (Morand et al. 2010). Importantly, this perceptual bias could not be accounted for by global low-level visual properties as these were normalized across stimuli. It is therefore very likely that the reflexive response to faces, which is beyond the control of the observer, is influenced by high-level visual properties.
Here, we quantify this preferential orienting response toward faces and, for the first time, link it to the timing of activity in the human brain. We used stimulus-locked event-related potentials (ERPs) to investigate, with high-temporal resolution, the neural dynamics preceding the occurrence of voluntary and reflexive saccades to faces and nonface visual objects. We employed a task-cued paradigm for which the stimulus type was task irrelevant as participants were not asked to perform a categorization. We aimed to determine at what stage in the saccadic programming, a voluntary saccade dissociates from a reflexive saccade, and more importantly, to unravel just “when” face sensory information impacts on the saccadic programming. We show electrophysiological evidence of an implicit categorization of faces in reflexive orienting responses with an early face-specific target selection at 40–60 ms over posterior parietal and parieto-occipital scalp regions. Most importantly, this early activity predicted the speed of the upcoming saccades toward the faces. This supports 1) controversial claims that the stimulus-response integration for a face starts at a very early stage of saccadic programming and 2) a 2-stage process in the representation of a face in involuntary spatial orienting with an initial, rapid implicit processing of the visual properties of a face, followed by subsequent stimulus categorization depicted by the N170 component.
Materials and Methods
Twelve healthy individuals (aged 20–37; mean = 24 years; 8 women) received financial compensation for participating in this study. Eleven of the 12 participants were right handed (Oldfield 1971). All reported normal or corrected-to-normal vision. All participants provided written informed consent to participate in this experiment, which was approved by the local Ethics Committee of the University of Glasgow.
Full details about the antisaccade paradigm and stimuli used in this experiment can be found in Morand et al. (2010). Briefly, neutral Western Caucasian faces (6 male and 6 female faces), cars (n = 12 from the database of Schweinberger et al. 2007), and noise phase scrambled patterns (n = 12) were used as stimuli. Faces and cars (image size: 128 × 128 pixels, 8 bits/pixel) were front view gray scaled photographs pasted onto a uniform gray background and subtended a visual angle of approximately 4.20° × 3.8° (Fig. 1A). Stimuli were presented on a gamma corrected 21-in. A Sony GDMF520 CRT monitor with a resolution of 1280 × 1024 and refreshed at 85 Hz. The monitor was placed at 68 cm from the chinrest.
To control for low-level visual properties between stimuli, faces, and car, images were equated for spatial frequency, luminance, and contrast. To this end, the average amplitude spectrum of all images in the dataset was calculated first and the phase of each image was then combined to this average amplitude spectrum. As a result, faces and cars were normalized for their amplitude spectra. Noise patterns were generated by randomizing the phase of the normalized face and car images. Finally, the root-mean-square contrast was also normalized for all images resulting in stimuli normalized for their “global” low-level visual properties.
Task and Procedure
Participants were instructed to generate a saccade either in the direction of the stimulus appearing on the screen (prosaccade) or in the opposite direction away from the stimulus (antisaccade; Fig. 1B). Each trial consisted of 4 stages. (1) The initial display consisted of a central fixation cross (0.8°) on a white background and lasted at least 1000 ms, while the experimenter initiated the stimulus presentation as soon as the participant's eyes were stabilized on the fixation cross. (2) This was followed by the display for 200 ms of a central cue (0.8° in size), whose color instructed the participant to generate either a prosaccade (green dot) or an antisaccade (red dot). (3) The cue display was followed by a 100-ms blank screen that was inserted before the presentation of the stimulus to reduce reactions times for both pro- and antisaccades and to further increase the difficulty in antisaccades (Munoz and Everling 2004). (4) The blank screen was followed by the stimulus display, which remained in view for 1000 ms. Stimuli (faces, cars, and noise patterns) appeared in each trial at 1 of the 2 possible peripheral locations on the horizontal meridian, at 10° either to the left or right of the center of the screen. Cue type (pro- or antisaccade), stimulus type (face, car, and noise pattern) and stimulus location (Left [LVF] or Right [RVF] visual fields) were randomly interleaved. Participants completed 12 blocks of 60 trials (720 trials in total) and were asked to perform the correct eye movement as rapidly and as accurately as possible. In this task-cued paradigm, the stimulus type was task irrelevant as participants were not asked to perform any categorization. Moreover, participants could not predict what type of stimulus nor what location would be presented on the next trial.
Saccadic Eye Movement Acquisition and Analysis
The horizontal and vertical positions of the eyes (left or right) were recorded with a video-infrared eye tracker (EyeLink 2K SR Research Ltd., Mississauga, Canada), which has a sampling rate of 1000 Hz and a spatial resolution of 0.01°. The system uses the center of the pupil and corneal reflection technique to define the pupil position. Participants were seated at the table with their head resting on a chin rest. Each of the 12 blocks started with a 9-point grid calibration and validation procedure to ensure accurate eye tracking. Participants were asked to saccade to a gray, circular disk that appeared sequentially (but unpredictably) in a 3 × 3 grid. Prior to testing, participants received 30 practice trials.
Data were analyzed off line with the Data Viewer Software (SR Research Ltd., Mississauga, ON, Canada). Saccades were detected using velocity and acceleration criteria of 30°/s and 8000°/s2. Only the first saccade after stimulus display, onset was analyzed. Trials were discarded if (1) the saccade latency was shorter than 80 ms (anticipations) or (2) the amplitude of the first saccade was less than 2° or (3) initial fixation was larger than 0.5° away from the fixation cross. Furthermore, blinks were excluded. These criteria led to an exclusion of an average of 5.9% of trials per subject. Discarded trials were equally distributed across conditions and stimuli. The resulting saccades belonged to 1 of the 12 conditions defined by saccade type (pro- and antisaccades), saccade direction (left and right visual fields), and stimulus type (face, car, and noise pattern).
EEG Acquisition and Analyses
Continuous electroencephalogram (EEG) was acquired at 1024 Hz through a 128-channel Biosemi EEG system (Biosemi V.O.F., Amsterdam, The Netherlands) covering the entire scalp. Analyses were performed using Cartool (http://brainmapping.unige.ch). As our main purpose was to reveal electrophysiologically when the stimulus category impacted upon the saccade programming (specifically when saccades toward faces and saccades toward other visual objects differed), we performed stimulus-locked averaging and focused our analysis on presaccadic activity (Ptak et al. 2011). Peristimulus epochs of EEG starting 100 ms prior to and ending 250 ms after the onset of the stimulus, which satisfied the inclusion criteria defined by the eye movement analysis (see above), were averaged for each experimental condition and participant, with a ±100 µV artifact rejection criterion. Before group averaging, data from artifact electrodes of each participant were interpolated using a spherical spline interpolation (Perrin et al. 1987). Data were baseline corrected using the prestimulus period (−100 ms to 0), bandpass filtered (0.18–60 Hz and using a notch at 50 Hz), and recalculated against the average reference. Only correct pro- and antisaccades generated in response to faces, cars, and noise patterns are described in the Electrophysiological Results.
To disentangle the effects specific to the stimulus identity (face, car, and noise) from the effects related to the programming of the different eye movements (pro- and antisaccade), 2 distinct analyses were conducted: First, we searched for “task-related” effects and compared ERPs with pro-and antisaccades directed to the same visual field, regardless of stimulus type. Then, we searched for “stimulus-specific” effects and compared ERPs with face, car, and noise stimuli presented in the same visual field for each task condition (pro- and antisaccade analyzed separately).
For both analyses (hereafter referred to as “task-effect” and “stimulus-effect” analyses), we used a multistep analysis procedure that examines local and global measures of the electric field at the scalp (Murray et al. 2008). Briefly, it entails analyses of the response strength and response topography to differentiate effects due to the modulation in the strength of the responses of statistically undistinguishable brain sources from alterations in the configuration of these sources (resulting in a change of scalp topography), as well as latency shifts in brain processes between experimental conditions. These methods have been successfully used for analyzing EEG data from larger electrode sensor arrays and have been detailed extensively elsewhere (Morand et al. 2000; Michel et al. 2004; Murray et al. 2008; Cappe et al. 2010; Cocchi et al. 2011).
ERP Waveform Analysis
We identified periods of amplitude modulations by contrasting the ERPs involved respectively in the task-effect and stimulus-effect analyses from each electrode, as a function of time after the stimulus onset in a series of pairwise comparisons (t-tests) and 1-way analysis of variances (ANOVAs). The results of the analyses are presented as an intensity plot representing time, electrode location, and P-values at each data point. Only differences with P-values ≤0.05 (Bonferroni corrected by the number of electrodes −1) and extending over at least 15 ms were considered reliable (Guthrie and Buchwald 1991) and will be illustrated in the Results section.
Response Strength and Topographical Analyses
Amplitude modulations of ERP waveforms do not allow a distinction between activation of different networks (with different topographies of the electric field at the scalp) and modulation of similar networks. To determine whether amplitude modulations of ERPs result from a change in response strength (which would be consistent with a modulation in response gain) or result from a change in response topography (which would be consistent with a change in underlying brain sources), we applied 2 types of analyses focusing on 2 features of the electric field. The first type of analysis identified the periods of time when the strength of the ERP differed across experimental conditions, irrespective of the topography of the ERP. The second type of analysis identified the periods of time when the topography of the ERP differed within and across experimental conditions, irrespective of the strength of the ERP.
Changes in electric field strength were determined by analyzing the global field power (GFP; Lehmann and Skrandies 1980). A GFP is calculated as the root-mean-square across all electrodes and represents the spatial standard deviation of the electric field at the scalp. Large values of GFP yield stronger electric fields. GFP was analyzed in a similar manner to what we described above for the analysis of ERP waveforms, using millisecond-wise paired t-tests on GFP waveforms (P-values ≤0.05 over at least 15 ms were considered). Observation of a GFP modulation without accompanying topographical changes across experimental conditions is best explained by a strength or power modulation of statistically undistinguishable neural generators across experimental conditions.
To identify statistically significant periods of topographical changes, we applied a spatio-temporal analysis approach that searches for topographical differences of the global electric field (potential map) between conditions across time (i.e., Morand et al. 2000; Schnider et al. 2002; Nahum et al. 2009). We emphasize that changes in the electric field topography derive from changes in the configuration of the brain's underlying active generators (Lehmann 1987). This method is based on a modified spatial k-means clustering analysis (Pascual Marqui et al. 1995) that determines the most dominant map topographies and the periods during which they are present in the whole group-averaged data. This approach is based on the observation that scalp topographies do not change randomly, but rather remain stable for a period of time in a certain configuration and then rapidly switch to a new configuration. The periods of stability have been called “functional microstates” (Lehmann 1987; Murray et al. 2008) and are thought to reflect the different information processing steps. Note that the topographical analysis method is independent of the reference electrode and is insensitive to amplitude modulation of the same scalp configuration across conditions, as topographies of normalized maps are compared. The optimal number of maps explaining the averaged data sets was determined with the cross validation (Pascual Marqui et al. 1995) and the Krzanowski–Lai criterion (Krzanowski and Lai 1985).
The cluster analysis was applied to the group-averaged ERPs involved in the task-effect analysis (pro- and antisaccades directed to the same visual field, all stimulus types pooled together) and in the stimulus-effect analysis (face, car, and noise presented in the same visual field separately for pro- and antisaccades). The distribution of the topographical maps across time was examined within and compared between conditions.
The significance of the pattern of maps observed in the group-averaged data was then tested statistically by comparing each of these maps with the moment-by-moment scalp topography of individual subject's ERPs. That is, each time point of the individual subject's ERPs was labeled according to the map with which it best correlated spatially, yielding a measure of map presence (global explained variance, GEV) that was in turn submitted to repeated-measures ANOVA using conditions (task or stimulus type) and map as within-subject factors. P-values of post hoc single comparisons were Bonferroni-corrected. This fitting procedure shows whether a given experimental condition is more often described by one map versus another and, therefore, whether different generator configurations better account for particular experimental conditions.
We assessed error rates for both the pro- and antisaccade performance for the 3 stimulus types (faces, cars, and noise) and the 2 stimulus locations (left and right). A 3-way ANOVA revealed main effects of task (F1,11 = 50.725, P = 0.00002), showing greater errors for anti- compared with prosaccades and of stimulus type (F2,22 = 14.237, P = 0.0001), showing greater errors for faces compared with other stimuli, yet there was no stimulus location effect. As the stimulus location did not have any effect on the independent variables, stimuli presented to the left and right visual field of the same task and stimulus type were pooled together. The resulting 2-way ANOVA revealed a main effect of task (F1,11 = 50.725, P = 0.00002), a main effect of stimulus type (F2,22= 14.237, P = 0.00011) and a significant interaction between task and stimulus type (F2,22 = 12.007, P = 0.0003). Pairwise comparisons showed that antisaccades generated higher error rates (20.65%) than prosaccades (1.76%, Fig. 2A,B). More importantly, face stimuli induced greater errors (24.12%) compared with cars (19.15%, F1,11 = 48.9, P < 0.001) and noise patterns (18.67%, F1,11 = 14.92, P = 0.0026; Fig. 2B). There was no difference in error rates for prosaccades for the 3 stimulus types (Fig. 2A).
Saccadic Reaction Times
Saccadic reaction times for correct pro- and antisaccades were analyzed in a 2-way repeated-measures ANOVA as described above. This analysis revealed a main effect of task (F1,11 = 17.2, P = 0.0016), a main effect of stimulus type (F2,22 = 5.3, P = 0.013) and a significant interaction between task and stimulus type (F2,22 = 7.2, P = 0.004). As expected, post hoc comparisons revealed that antisaccades (197 ms) were on average 26 ms slower than prosaccades (Fig. 2C,D). More interestingly, face stimuli induced faster prosaccades (166 ms) compared with cars (173 ms, F1,11 = 22.74, P < 0.001) and noise patterns (175 ms, F1,11 = 17.44, P < 0.001, Fig. 2C). No differences in latency were found for antisaccades (Fig. 2D).
The saccadic eye movement analysis thus revealed that faces induced both, a higher error rate in antisaccades and faster prosaccades in comparison with cars and noise patterns, replicating our previous study.
Task Effect (Pro- vs. Antisaccades)
Saccades directed to the left (LVF) and to the right (RVF) were analyzed separately. Figure 3 summarizes the results of the multistep analysis applied to pro- and antisaccades directed to the same visual field. As similar electrophysiological patterns of difference were found between eye movements directed to the left and right visual fields, except that the electrical field was flipped between left and right, we chose to present the differences found between pro- and antisaccades directed to the LVF only. We first identified electrode amplitude differences in the whole electrode set between ERP waveforms preceding prosaccades and antisaccades, by computing point-wise t-tests over the epoch covering −100 ms to +250 ms following the stimulus onset. These tests identified significant and temporally sustained differences (i.e., P < 0.05 for at least 15 ms) at 70–90 ms over central electrodes and from 110 ms to around 160 ms at central and right posterior electrode sites before extending over time to the left and right frontal electrodes up to 200 ms (Fig. 3A). It is worth noticing that differences in neural responses found after 171 ms (average latency for prosaccades, ±17 ms) most probably reflect differences in saccade latencies between pro- and antisaccades and are thus likely to be contaminated by eye movement execution. With this in mind, we think that it is very unlikely though that any differences found before 150 ms are linked to such eye movement artifacts and would argue that they are better explained by differences in the programming of saccadic eye movements.
The analysis of the GFP revealed statistically reliable differences in response strength of the saccade programming between 70–87 and 109–126 ms after the stimulus onset (Fig. 3B). The late time period of difference in response strength between pro- and antisaccades was concomitant with a topographical change at 110 ms, as identified by the topographic pattern analysis. Four stable electrical brain configurations that explained 94.3% of the variance (GEV) were identified over the poststimulus period. The temporal succession of these maps and their spatial distribution are shown in Figure 3C. While topographies were identical over the first 110 ms (map A) and after 200 ms (map C), 3 maps (maps B, D, and E) were identified during the 110–200-ms period that differentially accounted for pro- and antisaccades. The fitting procedure in individual's ERPs and within-subjects ANOVA over the 110–200-ms poststimulus period revealed a significant task × map presence interaction (F2,22 = 22.47, P < 0.0001) with maps D and E better characterizing the programming of antisaccades (map D: F1,11 = 25.28, P = 0.0003; map E: F1,11 = 7.6, P = 0.018) and with map B significantly better explained in the programming of prosaccades (F1,11 = 12.44, P = 0.0004). In contrast, the early difference in response strength found at 70–87 ms (Fig. 3B) was not temporally coincident with a change in topographical patterns. Map A was equally represented in pro- and antisaccades (GEV, F1,11 = 1.3495 , non significant [ns]), but prosaccades generated stronger global electrical response in this time window, suggesting a higher response magnitude of identical underlying brain sources for prosaccades compared with antisaccades.
To summarize, these results show that the neural response differences in the programming of pro- and antisaccades involve first a change in response strength of identical brain generators indicating a modulation of response gain between 70 and 90 ms for prosaccades, followed by a change in response topography at 110 ms consistent with different configurations of underlying brain sources for pro and anti saccades.
Stimulus Effect (Face vs. Other Visual Categories)
To investigate when the stimulus type interferes with the saccade programming, ERPs to face, car, and noise pattern stimuli presented in the same visual field were compared for pro- and antisaccades separately. Among all tasks (pro, anti directed to the LVF and RVF), only prosaccades directed to the LVF showed significant neuronal response differences across stimulus categories (Fig. 4). A time point-wise 1-way ANOVA applied to the ERPs waveforms from each electrode revealed a main effect of stimulus type in 2 time windows, at 40–60 ms and 125–140 ms after the stimulus onset (Fig. 4C). Further amplitude modulations were found from 165 ms on, which correspond to the time period of the average latencies of prosaccades to face, car, and noise patterns and consequently, and might reflect the neuronal activity related to motor execution. Post hoc tests (paired-wise t-tests with P-values <0.05 after Bonferroni correction) indicated that, at 40–60 ms, faces evoked stronger amplitude responses over right posterior parietal and parieto-occipital electrodes, compared with noise patterns (CP2, P2, P4, PO4,PO6, and PO8) and cars (P4 and PO4) as illustrated in Figure 4D (left panel). In the late time period of significant amplitude differences (125–140 ms), faces and cars generated stronger activity over parieto-occipital electrodes (PO6 and PO8) when compared with noise, respectively, while no significant differences were found between face and car stimuli (Fig. 4D, right panel). This suggests that the early stimulus effect at 40–60 ms over right parietal and parieto-occipital sites is “face specific”, whereas, at 125–140 ms, the stimulus effect seems to be object specific. To strengthen this finding, we performed a conservative correlation analysis between the mean amplitude response of each above-mentioned electrodes depicting the stimulus effect and the saccadic reaction time in the 2 time periods of interests. The early face-specific effect was confirmed by a direct relationship between the amplitude response for faces over PO6 and PO8 electrodes at 40–60 ms and the speed of saccadic execution. Indeed, early face-specific modulation over right parieto-occipital sites correlated negatively with the saccadic reaction time (nonparametric Spearman, R = − 0.64, P < 0.05), indicating that between 40 and 60 ms after the stimulus onset, the stronger (the more positive) the neuronal response to face stimuli, the faster the prosaccade execution (Fig. 5). No significant correlation could be established for faces during the late time period, nor for the car or noise patterns in any of the 2 time periods of interest. Finally, to exclude the possibility that the face-specific effect reported here at 40–60 ms could be due to a difference in reaction times to faces compared with cars or noise, we searched for latency differences in ERPs over the right parietal and parieto-occipital electrodes depicting the stimulus effect (CP2, P4, and PO8 as illustrated in Fig. 4D). No significant differences in latency were found across faces, cars, and noise patterns on electrodes CP2 (F2,22 = 1.32, P = 0.28, ns), P4 (F2,22 = 0.88, P = 0.42, ns), and PO8 (F2,22= 1.38, P = 0.27, ns). The latency of the maximal amplitude was on average 49 ms ±8.6 for faces, 48.11 ms ±8.8 for cars, and 49.6 ms ±9.4 for noise patterns over the CP2, P4, and PO8 electrodes. This demonstrates that the stimulus effect (e.g., face specific) in the parieto/parieto-occipital electrodes is not confounded by differences in reaction times between the different stimulus categories.
It is important to note that the GFP analysis failed to reveal statistically reliable differences in response strength (Fig. 4A). The topographic pattern analysis identified 3 stable electrical brain configurations over the poststimulus period that explained 96.2% of the variance of the group-averaged ERP dataset (Fig. 4B). The fitting procedure in individual's ERPs and within-subjects ANOVAs over the 0–170-ms poststimulus period revealed no significant stimulus type × map presence interaction or significant main effects, indicating that the 3 maps were equally explained in the prosaccades directed to face, car, and noise pattern stimuli. The early neuronal response modulation elicited by the face stimuli was not temporally synchronous with a change in topographical patterns across stimulus categories, suggesting a strength or power modulation of identical underlying brain sources across the contrasted categories.
We pinpoint the neural correlates associated with the fast involuntary orienting responses to faces described previously (Gilchrist and Proske 2006; Morand et al. 2010). We first report effects of task in the 70–90 and 110–200-ms time windows following target presentation. In addition, in the context of prosaccades, we observed an effect of stimulus as early as 40 ms following the target onset, with faces eliciting higher activity over parieto-occipital sensors. To disentangle face-specific effects that can occur at the level of stimulus processing from those at the level of motor programming during saccadic implementation, we first isolated the motor programming-level neural activity by comparing pro- with antisaccades.
Modulation in Amplitude and Topography During Pro- and Antisaccades
The results showed that the programming of pro- and antisaccades generated a first change in electrophysiological response strength for prosaccades at 70–90 ms after the stimulus onset over central electrodes. There were no accompanying topographical changes, indicating a modulation of response gain for prosaccades. This was followed by a change in response topography at 110 ms for antisaccades, with 2 successive electric field configurations, consistent with a modification of the underlying brain sources. These results are in line with a recent ERP study showing an increase in P1 component amplitude (onset at 70–90 ms) prior to reflexive saccades in a visual search task (errors in this case), which the authors interpreted as the neural correlate of a sensory gain control mechanism that enhanced target saliency or decreased distracter saliency through a modulation of sensory information during saccade planning (Ptak et al. 2011). Our data are in agreement with this interpretation as in antisaccades, the observer has to look away from the target and target selection and relevance might thus be reduced. The gain response observed at 70–90 ms in the reflexive orienting response (prosaccades) could possibly reflect stronger sensory signals for an enhancement of target selection. The fact that this was depicted over central electrodes supposes a possible role of the FEFs in this modulation activity. Several studies have shown that the FEF contributes to saccade target detection through top-down signals that modulate activity of visual neurons and, thus, target saliency (Moore and Armstrong 2003; Treue 2003). Moreover, in humans, the FEF exhibits very early visual activation in response to visual stimuli (Kirchner et al. 2009; Liu et al. 2009).
Our results further revealed distinct patterns of electrophysiological activity associated with voluntary and reflexive saccades at 110 ms after the stimulus onset, on average about 80 ms prior to the onset of the saccade. Antisaccades were characterized by 2 successive topographical maps (110–170 and 170–200 ms) that were concomitant with a modulation of the amplitude response over the central and parietal electrodes. These additional processes in antisaccades could possibly reflect the top-down inhibition of the reflexive, automatic prosaccade and the reprogramming of a voluntary saccade. In fact, previous EEG studies have reported differences in a similar time range between the programming of reflexive and voluntary saccades and have identified the FEF, IPS, and the DLPFC as good candidates for the source of the top-down inhibitory control and saccade re-programming (i.e., Everling et al. 1998; Mc Dowell et al. 2005; Clementz et al. 2007; Ptak et al. 2011). Interestingly, a transient presaccadic increase in alpha range brain oscillation (7–8H z) has been observed over central–parietal electrodes in voluntary saccades starting 100 ms prior to the onset of the saccade and was interpreted by the authors as a marker of suppression of bottom-up oculomotor capture (Mazaheri et al. 2011).
Early Target Selection to Faces in Prosaccades
Most interestingly, our study revealed a very early gain response over right posterior parietal and parieto-occipital sites for prosaccades. This response modulation was specific to faces and occurred at 40–60 ms following the onset of the stimulus. Crucially, it was also predictive of the speed of the saccade execution: The greater the amplitude over parieto-occipital sites the faster the prosaccade toward faces. This is evidence for an early target selection to faces in reflexive orienting responses over parieto-occipital cortex that facilitates the triggering of saccades toward faces. Moreover, numerous studies have shown hemispheric asymmetries for faces with preferential right over left hemisphere processing (De Renzi et al. 1994; Wojciulik et al. 1997) and, more specifically, an LVF advantage for face processing (Burt and Perrett 1997; Luh et al. 1991; Dutta and Mandal 2002; Schyns et al. 2002; Butler et al. 2005; Butler and Harvey 2006, 2008; Yovel et al. 2008). These studies fit perfectly with our finding of the early gain response occurring over the right hemisphere (for targets in the LVF).
It is important to clarify that this finding does not contradict the numerous studies demonstrating the link between the N170 and face categorization (e.g., Schyns et al. 2007; Van Rijsbergen and Schyns 2009; Vizioli et al. 2010) as our paradigm did not require categorization. Participants were simply asked to move toward or away from an upcoming target, and the target contents were irrelevant to the task. The early face effect described here for the prosaccades might instead indicate a multiphase process in the representation of a face in automatic spatial orienting, implying a rapid preselection for faces in comparison with other nonface categories, followed by subsequent stimulus categorization depicted by the N170. Our data support that this hypothesis as a significant response amplitude was found at 120–140 ms differentiating between stimulus categories and noise patterns, a latency compatible with the onset of the N170 component.
The rapid selection of faces in the reflexive saccades is most probably due to the processing of the visual properties of the stimulus, rather than to an effect of the faces on the motor programming of the saccadic eye movements. We argue that this as there was no overlap in time between the face-specific modulation identified at 40–60 ms over parieto-occipital sites in prosaccades and the effects due to the programming of pro- and antisaccades (identified at 70–90 ms over central sites for all stimuli categories). Furthermore, the early selection of faces cannot be accounted for by the global low-level visual properties of the faces as these were normalized across all stimulus categories. This ruled out the possibility that the faces could be discriminated solely on the basis of a bias in the global contrast, luminance or spatial frequency content or, indeed, a difference in the spectral profiles of the contrasted categories (VanRullen and Thorpe 2001; VanRullen 2006; Honey et al. 2008; Schyns et al. 2003). Thus, we think that the early selection for faces could either be driven by high-level “social” properties (e.g., the global appearance of a face, familiarity, expertise) or/and local low-level physical properties intrinsic to faces (e.g., the high contrast present in the eye region, spatial frequencies carried by face features). In any case, high-level or local low-level visual properties, or their combination, render face stimuli more salient than other visual objects and this could trigger their early selection.
Our results bring a possible electrophysiological explanation to recent studies showing fast saccadic detection/categorization for faces in complex scenes in just 100 ms, both in humans and monkeys (Crouzet et al. 2010; Girard and Koenig-Robert 2011). The fast target selection to faces in involuntary orienting responses in the parieto-occipital scalp region reported here, together with the extreme rapidity of saccades toward faces in saccadic detection/categorization tasks (Crouzet et al. 2010; Girard and Koenig-Robert 2011) places constraints on the brain mechanisms underlying the processing of complex stimuli. It implies a fast integration between the structures involved in the control of eye movements and spatial attention, among those the FEF and the posterior parietal cortex of the dorsal stream, the superior colliculus, and the face fusiform area (FFA), part of the ventral stream and responsible for face processing. The distribution of neuronal latencies in different cortical areas (Nowak and Bullier 1997; Schmoleski et al. 1998) precludes the possibility that the saccadic categorization process uses multiple iterations between brain regions before the motor response. Visual information has been shown to rapidly reach the human FEFs in very short latencies (Blanke et al. 1999), with only 45 ms in response time to natural scene containing faces (Kirchner et al. 2009). These latency data fit our results over the posterior parietal and parieto-occipital scalp regions. A rapid stimulus-driven enhancement in response discriminability has been described in ventral area V4 quasi-instantaneously after microstimulation of FEF (Amstrong and Moore 2007). A growing body of TMS studies has demonstrated a role for FEF in modulating extrastriate cortex activity with respect to the target (Grosbras and Paus 2003; Silvanto et al. 2006; Taylor et al. 2007), in as early as 40 ms (O'Shea et al. 2004; Morishima et al. 2009). Moreover, reduced category selectivity to faces and scenes over the extrastriate cortex has been described recently following a disruption of the PFC by repetitive transcranial magnetic stimulation (Miller et al. 2011). It is thus conceivable that the FEF, with its sensitivity to faces, enhances the representation and selection of faces through top-down modulating responses over extrastriate cortex during spatial attention.
Given the absence of electrical source localization analysis in this study and the low spatial resolution of the EEG technique, we can only speculate about the exact cortical areas activated during the early face-specific ERP modulation at 40–60 ms over the right posterior parietal and parieto-occipital scalp sensors in reflexive saccades. Nevertheless, this early modulation fits perfectly with crucial results reported by Sadeh et al. (2011). For the first time, they demonstrated a causal relationship between the activation of category selective (face and body) areas and ERP to these preferred categories, showing that early TMS stimulation (60–100 ms after stimulus onset) to subjects' individually defined right occipital face area (OFA) significantly increased the subsequent N170 amplitude to faces. Complementary to this, we now report an early face-specific modulation over similar (parieto-occipital) sites.
Moreover, Sadeh et al. (2010), in a combined EEG-fMRI study found a high correlation between face selectivity in the OFA and ERP selectivity around 100 ms, whereas the N170 face selectivity was not correlated with OFA face selectivity but instead correlated with the FFA and pSTS (posterior superior temporal sulcus) activity. Thus, they argue that, in their experiment, the magnetic stimulation modulated neural output from the OFA to face-selective areas in the temporal cortex, which in turn directly contributed to the N170 component (Sadeh et al. 2011). Apart from our early face-selective modulation, we also found an N170 modulation on similar scalp sites as described in Sadeh et al. (2010, 2011), yet in our study, this modulation was object- but not face-specific. It is possible that our analyses lacked the power to pick up these sites and/or that, due to the normalization procedure, applied to exclude low-level visual confounds across stimulus categories, our car stimuli might have been too similar to the face stimuli to allow a distinction at this later component. Nevertheless our data agree with Haxby et al.'s (2000) model, which proposes feed-forward connections within the core face processing system from the OFA to both the FFA and pSTS. More specifically, we suggest a 2-stage process in the representation of a face in involuntary spatial orienting with an initial, rapid implicit processing of the visual properties of a face (either social, high-level properties and/or physical local low-level visual properties), not driven by global low-level visual properties, followed by subsequent stimulus categorization depicted by the N170 component.
To conclude, we bring here electrophysiological evidence of an early target selection in reflexive orienting to faces, occurring over parieto-occipital sites at a very early stage of saccade programming. We suggest that this fast selection for faces is related to the implicit processing (or preprocessing) of the visual properties of a face not driven by global low-level visual properties.
We thank Gregor Thut for helpful discussion and Roberto Caldara and Luca Vizioli for previous analyses. This work was supported by a grant from the Economic and Social Research Council and the Medical Research Council (RES-060-25-0010) awarded to all authors. Conflict of Interest: None declared.