Cerebral Cortex V 14 N 6 �� Oxford University Press 2004 all rights reserved Cerebral Cortex June 2004 14:619���633 DOI: 10.1093/cercor/bhh023 Electrophysiological Correlates of Rapid Spatial Orienting Towards Fearful Faces Gilles Pourtois1, Didier Grandjean2, David Sander2 and Patrik Vuilleumier1,3 1Neurology & Imaging of Cognition, University of Geneva, Switzerland, 2Geneva Emotion Research Group, University of Geneva, Switzerland and 3Department of Psychology, University of Geneva, Switzerland We investigated the spatio-temporal dynamic of attentional bias towards fearful faces. Twelve participants performed a covert spatial orienting task while recording visual event-related brain potentials (VEPs). Each trial consisted of a pair of faces (one emotional and one neutral) briefly presented in the upper visual field, followed by a unilateral bar presented at the location of one of the faces. Participants had to judge the orientation of the bar. Comparing VEPs to bars shown at the location of an emotional (valid) versus neutral (invalid) face revealed an early effect of spatial validity: the lateral occipital P1 component (~130 ms post-stimulus) was selec- tively increased when a bar replaced a fearful face compared to when the same bar replaced a neutral face. This effect was not found with upright happy faces or inverted fearful faces. A similar amplifi- cation of P1 has previously been observed in electrophysiological studies of spatial attention using non-emotional cues. In a behav- ioural control experiment, participants were also better at discrim- inating the orientation of the bar when it replaced a fearful rather than a neutral face. In addition, VEPs time-locked to the face-pair onset revealed a C1 component (~90 ms) that was greater for fearful than happy faces. Source localization (LORETA) confirmed an extra- striate origin of the P1 response showing a spatial validity effect, and a striate origin of the C1 response showing an emotional valence effect. These data suggest that activity in primary visual cortex might be enhanced by fear cues as early as 90 ms post-stimulus, and that such effects might result in a subsequent facilitation of sensory processing for a stimulus appearing at the same location. These results provide evidence for neural mechanisms allowing rapid, exogenous spatial orienting of attention towards fear stimuli. Keywords: emotion, face perception, fear, human electrophysiology, source localization, spatial attention Introduction Different lines of evidence suggest that threat-related signals are rapidly and efficiently processed by specialized emotion mechanisms (LeDoux, 1996 ��hman and Mineka, 2001 Adolphs, 2003) and that our attention tends to be prioritized towards threat rather than neutral stimuli (Fox, 2002 Vuil- leumier, 2002). Behavioural studies have used a variety of para- digms borrowed from cognitive psychology to explore the effects of emotion on spatial attention, including covert orienting in dot-probe tasks (Mogg and Bradley, 1999 Mogg et al., 2000), visual search (Hansen and Hansen, 1988 Fox et al., 2000 Eastwood et al., 2001 ��hman et al., 2001), stroop inter- ference (Pratto and John, 1991 Williams et al., 1996) and atten- tional blink experiments (Anderson and Phelps, 2001). Most of these studies found that negative or threat-related stimuli may summon attention more readily than neutral stimuli. Thus, people are quicker at detecting fearful or angry faces among neutral distracters than vice versa (Hansen and Hansen, 1988 Fox, 2002), quicker at identifying probes replacing the loca- tion of threatening rather than neutral faces or words (Mogg et al., 1997), and better at perceiving words with aversive meaning than neutral words (Anderson and Phelps, 2001). Such attentional biases might be particularly pronounced in anxious individuals as compared with matched non-anxious control subjects (e.g. Fox, 1993, 2002 Mogg et al., 1994). Simi- larly, studies in brain-damaged patients with impaired spatial attention and hemi-neglect have shown that their detection of stimuli in the contralesional visual field is better for emotional pictures (Vuilleumier and Schwartz, 2001a) or emotional faces (Vuilleumier and Schwartz, 2001b Fox, 2002) than for neutral stimuli with similar visual complexity. Consistent with these behavioural findings, brain-imaging results in normal volunteers have revealed increased responses to threat-related pictures (e.g. fearful faces) in several areas of visual cortex, in addition to limbic regions such as the amygdala (Vuilleumier et al., 2001 Pessoa et al., 2002). These increases are thought to reflect enhanced attention towards emotional stimuli (Lane et al., 1998 Vuilleumier, 2002). However, the exact time-course and neural bases of attentional orienting towards emotional stimuli has yet to be determined. The current study used evoked potentials and source localiza- tion methods with the aim of identifying electrophysiological correlates of emotional biases in attention on a millisecond scale, and comparing these emotional effects with the results of previous studies manipulating spatial orienting with neutral cues (Clark and Hillyard, 1996 Hillyard and Anllo-Vento, 1998). A classical paradigm extensively used to examine both behavioural and neurophysiological effects of spatial attention is derived from Posner���s covert orienting task (Posner et al., 1980 Navon and Margalit, 1983), in which a target stimulus is preceded by a brief cue correctly predicting the location of the target (valid cue) or incorrectly predicting another location (invalid cue). Evidence for involuntary, reflexive, exogenous orienting is demonstrated by a facilitation of stimulus processing after valid cues and an interference after invalid cues, typically arising with stimulus onset asynchronies as short as 100 ms post-cue (Jonides, 1981 Egeth and Yantis, 1997). Variants of this paradigm have been used in behavioural (e.g. Fox, 1993 Bradley et al., 1997) and brain-imaging studies (Armony and Dolan, 2002) that have examined emotional influences on spatial attention. Thus, when a peripheral dot-probe is presented randomly in either the right or left visual field, preceded by a brief display with an emotional stimulus at one location and a neutral stimulus at the other location, subjects are quicker and/or more accurate at making judgments on the dot-probe if it appears on the same side as an emotionally nega-

620 Rapid Spatial Orienting Towards Fearful Faces ��� Pourtois et al. tive stimulus (a ���valid��� cue), rather than on the opposite side at the location of a neutral stimulus (an ���invalid��� cue). In other words, visual selection of the probe is facilitated by the emotional value of the preceding visual stimulus, based on the common spatial location. A recent functional magnetic reso- nance imaging (fMRI) study using a similar task (Armony and Dolan, 2002) also demonstrated that target probes presented at the location of neutral face paired with an aversively (sound) conditioned face at another location elicited a stronger activa- tion in fronto-parietal areas implicated in spatial attention in the latter as compared with the former condition, suggesting an involuntary capture of attention by the aversive stimulus that required subsequent reorienting towards the probe at the neutral location. Variants of Posner���s paradigm have also been used exten- sively in visual event-related potential (VEP) studies of spatial attention but employing simple non-emotional visual stimuli, such as chequerboards or gratings. Very consistent observa- tions have been obtained via such electrophysiological work over the last 20 years (Luck, 1995 Mangun, 1995 Clark and Hillyard, 1996 Eimer, 1998 Luck et al., 2000 Martinez et al., 2001). In many of these VEP studies (see Hillyard and Anllo- Vento, 1998), spatial attention was cued towards one visual field (e.g. endogenously by central cues), while bilateral grating stimuli were presented with a target appearing on the side that was either correctly cued (valid trials, e.g. 70%) or incorrectly cued (invalid trials, e.g. 30%). The typical results indicate that (i) selective spatial attention can produce early effects on the response to peripheral visual stimuli (within 200 ms post-onset) (ii) these effects are mainly manifested on the scalp as an increased amplitude of exogenous visual compo- nents (P1 and N1 waves), with greater responses on valid versus invalid trials (but no changes in latency or topography) (iii) the neural sources of these effects take place in extrastriate visual cortex, presumably corresponding to enhanced sensory processing. Depending on the task, attentional effects on the P1 component can be dissociated from those on the N1, with the latter being less modulated during bilateral than unilateral visual stimulation (Heinze et al., 1990 Luck et al., 1990 Lange et al., 1999) and more sensitive to attentional manipulations demanding feature discrimination rather than detection (Luck et al., 1990 Mangun and Hillyard, 1991 Vogel and Luck, 2000 Hopf et al., 2002). Another consistent finding has been that electrical activity of the primary visual cortex, indexed by the C1 component, does not seem to be involved in spatial atten- tion within this initial time-range of visual responses (Martinez et al., 1999), although primary visual cortex might be modu- lated at a later delay through feedback mechanisms from higher cortical areas (Martinez et al., 1999 Noesselt et al., 2002). To our knowledge, no VEP study has directly investigated similar neurophysiological indices of attention using threat- related stimuli (e.g. fearful faces) in such a classical paradigm. Previous VEP studies have always presented emotional faces centrally, at an attended location (see Halgren et al., 2000 Campanella et al., 2002 Pizzagalli et al., 2002 Eger et al., 2003), although one study tested for effects of spatial attention on the response to emotional stimuli presented in the peri- pheral visual field (Holmes et al., 2003), and another study presented facial expressions unilaterally in each hemifield to examine hemispheric asymmetries (Pizzagalli et al., 1999). However, no study has directly tested for the effects of emotional cues on spatial attention. This was the aim of our current study, by using high-density EEG recording and source localization in normal observers during covert spatial orienting in a dot-probe task, where emotional and neutral faces served as valid and invalid cues, respectively. We used a typical version of this task, adapted from Mogg et al. (1994). On each trial, two faces were briefly presented, one in each visual field, one neutral and one with an emotional expression (fearful or happy). The two faces were then replaced by a small bar-probe at the position just occupied by one of them, oriented either vertically or horizontally (Fig. 1). Partici- pants were asked to judge the bar orientation as quickly as possible. The bar unpredictably appeared on the side of the emotional face (valid condition) or on the side of the neutral face (invalid condition), but importantly, both neutral faces and emotional expressions were entirely irrelevant to the participants��� task. Only short time intervals (100���300 ms) between the face pair and the bar onset were used in order to tap exogenous mechanisms of spatial orienting (Jonides, 1981 Egeth and Yantis, 1997). In comparison with previous behav- ioural studies using emotional dot-probe tasks, we introduced three important methodological changes. (i) The two faces in the pair were always of two different individuals in our para- digm, whereas the majority of previous studies used faces with the same identity but either the same or different expressions. The former design makes it easier to disentangle attentional biases due to image differences or true emotion-specific effects, since emotional expression is the only facial dimension systematically associated with the spatial validity manipulation, rather than other changing or deviant properties of a particular stimulus within the pair. (ii) Both faces and the bar probe were presented in the upper visual field in our study, such that they could elicit a robust C1 component in the EEG, reflecting early V1 activity, i.e. with a negative wave corresponding to retin- otopic responses for the upper field stimulation (Jeffreys and Axford, 1972a,b). Previous studies presented faces on the hori- zontal meridian, which would cancel out the upper and lower field components of C1 and make difficult to differentiate the C1 from the P1. By contrast, the peripheral presentation of our stimuli in upper quadrants enabled us to test whether any effects of attention or emotion would affect the primary visual cortex (Clark et al., 1995). (iii) We opted for a go/no-go matching task in which participants had to judge, on each trial, whether the orientation of the bar probe (in the left or right upper visual field) matched that of a thick line segment within the fixation cross. The task was to press a button only when the bar orientation was the same as the thicker line of the cross (rare go trials), but to withhold responses otherwise (more frequent no-go trials). This task ensured that participants main- tained their gaze on the central fixation cross and that all stimuli were indeed presented in the upper visual field while we recorded VEPs to the face pair and to the bar probe. More- over, the choice of a go/no-go paradigm rather than a simple detection response was backed up by behavioural studies showing that attentional biases towards threat stimuli can persist across a variety of different tasks (see Mogg and Bradley, 1999). A go/no-go detection task was also more appropriate to record VEP uncontaminated by any motor-related activity. Thus, in the current EEG experiment, only VEPs generated on the no-go trials were analysed. Our hypothesis was that the sensory processing of bar probes should be enhanced when replacing a fearful face, if

Cerebral Cortex June 2004, V 14 N 6 621 spatial attention was involuntarily oriented towards that partic- ular location, as compared with the location of a neutral face. Therefore, we expected that VEPs elicited by the bars should differ as a function of the spatial validity defined by the posi- tion of preceding faces. We had two main predictions. First, spatial validity should modulate the P1 component previously identified as a marker of selective focusing of attention during bilateral visual stimulation, more than the N1 component that is sensitive to other attentional conditions (Luck et al., 1990 Hopfinger and Mangun, 1998). Secondly, any biases in spatial attention might be either specific, stronger or faster for bars replacing fearful faces, as compared with happy faces, in keeping with behavioural data suggesting greater effects of negative than positive stimuli (see Fox, 2002 Vuilleumier, 2002). In addition, we also determined whether a behavioural effect of validity was obtained in our modified dot-probe para- digm, using a separate control experiment with the same stimuli and the same go/no-go task. Since our EEG experiment required a low number of go trials for VEPs uncontaminated by motor artefacts, and therefore provided few reaction time measures, our behavioural control experiment used a higher probability of orientation matching between the bar and the fixation cross as compared with the EEG experiment. Finally, we established that our attentional effects were truly driven by facial expression rather than low-level pictorial cues in another EEG control experiment using inverted faces. Materials and Methods Participants In the main EEG experiment, participants were 14 right-handed intro- ductory psychology and medicine students (nine female, with a mean age of 22 years, SD 2.5 years) from the University of Geneva. Two participants were excluded from statistical analyses because of exces- sive alpha band in the EEG contaminating the signal by occurring within the same frequency band (���10 Hz) as the VEPs of interest. Six other students (five female, mean age of 23 years, SD 1.6 years) parti- cipated in an EEG control experiment, and another 16 volunteers (13 female mean age 23 years, SD 2 years) who did not participate in the EEG experiments took part in the behavioural control experiment. All participants had normal or corrected to normal vision, and were free of neurological or psychiatric history. Materials The face stimuli were pairs of grey-scale photographs of ten different individuals (four males and six females), all taken from the stand- ardized Ekman series (Ekman and Friesen, 1976). Each face pair consisted of two different identities with the same gender, one portraying an emotional expression (fearful or happy) and the other a neutral expression. Four pair conditions were used: fear���neutral, neutral���fear, happy���neutral and neutral���happy (Fig. 1). Each emotion expression appeared equally often to the left or right of the neutral expression. Thus, thirty faces (3 emotions �� 10 identities) were used, and for each condition 30 pairs were constructed by combining each individual with three other individuals. Each face stimulus was trimmed to exclude the hair and non-facial contours, and enclosed within a rectangular frame measuring 8 �� 10 cm, subtending 6.5�� �� 8.2�� of visual angle at a 70 cm viewing distance (227 �� 285 pixels on a 256 grey-level scale). Each face stim- ulus was analysed in Matlab (Fig. 1) to extract the mean pixel lumi- nance, contrast range, surface size occupied by the face and value of central spatial frequency (Nasanen, 1999 Bex and Makous, 2002). Non-parametric analyses of variance on these measures revealed that neutral, fearful and happy faces did not differ in average pixel lumi- nance [Kruskal���Wallis test, H(2) = 2.99, P = 0.22], luminance contrast [Kruskal���Wallis test, H(2) = 0.22, P = 0.90], face size [Kruskal���Wallis test, H(2) = 0.41, P = 0.82] or central spatial frequency [Kruskal���Wallis test, H(2) = 3.91, P = 0.14]. All stimuli were presented on a black background, on a 17 in. computer screen with a PC Pentium 2 running Stim. The vertical position of the screen was adjusted for each subject so that the level of the fixation cross was at the horizontal meridian. A fixation cross meas- uring 2 �� 2 cm (thickness 0.1 cm) was presented centrally in the lower part of the computer screen. The faces were presented in the upper visual field at an eccentricity of 4.1��: the distance between the hori- zontal meridian and the outer edge of the face was 5 cm. The faces were equidistant from the vertical meridian, and each face centres were 18 cm apart (14.7��). The probe was a white rectangular bar (either horizontal or vertical) measuring 6 �� 0.4 cm (4.9�� �� 0.33��). It was presented on either the left or right side of the screen, its centre being 9 cm (lateral) Figure 1. (a) Procedure used in all our EEG and behavioural experiments, showing the sequence of events within a trial. (b) The four different face pair conditions (fear���neutral, neutral���fear, neutral���happy and happy���neutral) that served as exog- enous cues. The spatial distance and coordinates of the face relative to the fixation cross are shown. (c) The mean face images for fearful (left), happy (middle) and neutral (right) conditions as computed by overlapping and averaging all faces with the same expression (n =10). There was no conspicuous difference in low-level properties (e.g. luminance, size and spatial frequency) for the different emotional face conditions, as confirmed by further quantitative analyses (see Materials and Methods).