This event-related fMRI study investigated the impact of processing load on the neural response to social-emotional stimuli in healthy adults. Significantly enhanced amygdala activity to fearful relative to neutral facial expressions evident during gender judgments was absent during the unattended conditions that included processing load. Moreover, data from the ANOVA also revealed a significant main effect of processing load within the amygdala. Contrary to theories of automaticity, amygdala activity diminished as processing load increased. In addition, BOLD responses in medial prefrontal cortex and sensory representation areas of temporal cortex decreased as processing load increased. Conversely, increased processing load was associated with increased activity in dorsolateral prefrontal and parietal cortices. In accordance, connectivity analysis measuring correlated activity across all task conditions revealed robust correlated activity from dorsolateral prefrontal cortex to parietal and ventrolateral cortices. Furthermore, this activity was negatively correlated with activity in ventromedial prefrontal cortex. The data are consistent with notions that the processing of emotional information, like neutral information, is governed by top-down processes involved in selective attention.
We observed amygdala activity in response to composite stimuli during gender discriminations, but deactivation to the same stimuli when processing load increased. Theories of automaticity predict that the amygdala responds to emotional stimuli independent of attention, perhaps through a subcortical route (Morris, Ohman, & Dolan, 1998
; Morris et al., 2001
). However, the current data support the view of Pessoa and his colleagues (2002
, which suggests that emotional expressions are not a “privileged” category of object immune to the effects of attention. The fact that neural activity associated with social and emotional stimuli was reduced with increased processing load supports the notion that facial stimuli must also compete for neural representation. This pattern was observed not only in the amygdala, but also within medial prefrontal cortex and superior temporal sulcus. Both regions are implicated in social processing. Regions of medial prefrontal cortex have been implicated in emotional processing (Damasio, Tranel, & Damasio, 1990
; Rolls, 1996
), perhaps representing reinforcement information provided by the amygdala (Blair, 2004
; Kosson et al., 2006
). The superior temporal sulcus has been implicated in face processing (Chao et al., 1999
; Haxby et al., 2000
). Together, our results suggest that the response of the amygdala and medial frontal cortex to social-emotional stimuli is subject to top-down control related to attentional selection.
A previous study has shown that the level of processing load can be successfully modulated by task demands when participants engage in case and syllable discriminations of linguistic stimuli (Rees, Frith, & Lavie, 1997
). In our study, increased processing load was associated with increased activity in parietal cortices and a large area of prefrontal cortex extending from dorsolateral cortex to parietal cortex. These regions are implicated in selective attention, particularly posterior parietal cortex (Behrman, Geng, & Shomstein, 2004
; Nobre et al., 1997
; Nobre et al., 2004
; Posner & Peterson, 1990
) and dorsolateral prefrontal cortex (Dias et al., 1996
; Liu et al., 2004
In the current study, greater activity to fearful versus neutral faces in middle occipital gyrus was observed across conditions suggesting that some form of emotional processing occurred across task conditions. In addition, the error data (though not the reaction time data) showed a similar main effect of emotion. However, activity in the amygdala was reduced at high levels of processing load. One possibility is that a weaker signal originating in the amygdala was present at higher levels of processing load, but was not detectable via measures of BOLD response due to susceptibility of signal drop-out. Another possibility is that the signal was not detected because it was transient. A recent event-related potentials study showed that activity evident while attending to emotional facial expressions was also evident during inattention; however, during inattention, the response was transient, and extinguished within 220 ms (Holmes, Kiss, & Eimer, 2006
). In addition to restrictions placed by limited perceptual processing capacity, stimuli may be selected further through executive attention (i.e., goal-directed) mechanisms that augment the strength of target stimuli at the expense of irrelevant ones (Desimone & Duncan, 1995
). This raises the possibility that in the current study, rapid emotional responses may have occurred even at high levels of processing load, but were subject to cognitive modulation through executive attention mechanisms. For example, very early fear-differentiating amygdala activity may occur across task-conditions via a proposed “automatic” sub-cortical route (LeDoux, 1996; Morris et al., 1999
; Whalen et al., 1998
). Although sufficient to influence activity in occipital gyrus, this early differentiated amygdala signal might be disrupted by the competing cortical representation of the target stimuli during case and syllable discriminations. A recent study involving MEG suggests that the amygdala distinguishes fearful from neutral faces by 30ms following stimulus-onset (Luo et al., in press
). At least some animal data would support the existence of these conduction speeds (e.g., Quirk et al., 1995
). It should be noted that other work, involving intra-cranial recordings in humans and with more complex pictorial stimuli (scenes rather than faces), have suggested that amygdala activity distinguishes pleasant from unpleasant images between 50 and 150ms (Oya et al., 2002
). Additional studies involving methods with high levels of temporal resolution will help address these questions concerning the speed and susceptibility to regulation of amygdala activation.
A significant processing load by emotion interaction was observed in anterior cingulate cortex and ventrolateral prefrontal cortex. Both neural regions showed increased activity in the presence of fearful stimuli during the attended condition (gender discrimination) and during the high processing load condition (syllable discrimination), but increased activity to neutral versus fearful faces during the low processing load condition (case discrimination). It is difficult to interpret this effect with respect to valence given that the amygdala, which is the structure involved in emotional processing often considered most robust to attentional manipulations (e.g., Anderson et al., 2003
), was not significantly active at higher levels of processing load. Furthermore, activity in these regions was greater to neutral faces than fearful faces at low levels of processing load in the unattended condition. It is interesting to note that the two areas identified by the interaction are regions frequently implicated in reacting to response conflict and control (see Blair, 2004
; Botvinick et al., 2004
; Casey et al., 2001
; Luo et al., 2006
). However, it is unclear why these two regions should show increased activity to fearful relative to neutral expressions in the gender and syllable discrimination conditions but the inverse in the case discrimination conditions.
According to current models of conflict monitoring, cognitive (“attentional”) control is enacted in prefrontal cortex, particularly left dorsolateral prefrontal cortex (MacDonald et al., 2000
), in situations of increased conflict (Botvinick, Cohen, & Carter, 2004
), and in the presence of threatening distracters (Bishop et al., 2004
). With reference to this model, regulation of the emotional response to facial stimuli may have occurred through at least two pathways. During circumstances of greater conflict, regions of prefrontal cortex, particularly medial prefrontal cortex, may eliminate the competitive advantage afforded to emotional stimuli through direct modulatory connections with the amygdala. This functional relationship has been proposed on the basis of animal data (Quirk & Gehlert, 2003
), and imaging studies (Pezawas et al., 2005
). A second possibility is that regions of prefrontal cortex, particularly lateral prefrontal cortex, manipulate representations of task relevant stimuli in temporal regions at the expense of distracters (Botvinick, Cohen, & Carter, 2004
). For example, the modulation of the strength of auditory representations is thought to result from regulatory projections from dorsolateral prefrontal cortex to auditory processing regions in temporal cortex (Barbas et al., 2005
; Chao & Knight, 1997
). A similar mechanism may exist for object and face representation areas of temporal lobe either through direct projections between these regions and prefrontal cortex, or indirectly through parietal cortex. In addition to interactions between amygdala and prefrontal cortex, emotional regulation may also be achieved through mechanisms associated with attentional selection and manipulating the strength of object representations (Mitchell et al., 2006). Both processes may be active in the current study. In order to further investigate the potential contribution of cognitive control of object representations, we conducted a connectivity analysis.
Following considerable data implicating dorsolateral prefrontal cortex in enacting attentional control over competing stimulus representations (Botvinick, Cohen, Carter, 2004
; Liu et al., 2004
; MacDonald et al., 2000
), we selected the area of peak intensity within this region as our “seed” for the connectivity analysis. In line with our predictions, dorsolateral prefrontal cortex showed correlated activity with ventrolateral prefrontal cortex, and superior parietal regions, and was negatively correlated with activity in the ventromedial prefrontal cortex. Interestingly, this network of activity corresponds closely to areas observed in studies of emotional regulation in which subjects are asked to augment or reduce their emotional response to positive and negative pictorial stimuli (Ochsner, 2005
). We suggest that as in our study, emotional regulation may be facilitated by activity in prefrontal and parietal regions that manipulate the salience of task-relevant or “non-emotional” stimulus features in temporal cortex at the expense of representations of emotional information.
An important caveat should be noted with respect to the nature of the stimuli used in the present task. Although the linguistic and facial stimuli used were spatially overlapping, there is a possible confound between processing load and the spatial extent of attentional focus. The spatial location subject to attentional focus was smaller in the word-related task than in the gender discrimination task. Thus, some critical emotional cues, particularly the mouth, might fall outside the spatial location of attentional focus. As a consequence, reduced amygdala activation in the linguistic conditions might not be due to higher processing load, but rather to the narrower spatial focus of attention. In the study by Anderson et al. (2003)
, in which amygdala responses to attended and unattended fearful faces were the same, the spatial extent of overlapping objects had been kept constant. However, the main conclusions of our study are not affected by this potential confound; activity in neural regions associated with emotional or social processing (medial PFC and superior temporal gyrus) diminished with increasing processing load.
This event-related fMRI study provides further evidence for the notion that significant amygdala BOLD response to behaviorally peripheral social or emotional stimuli requires attention. Amygdala activity, like activity in medial prefrontal cortex and superior temporal sulcus, diminished when processing load increased. In contrast, activity in attention-related regions increased with increased processing load. Collectively, the data suggest that the processing of emotional information, like neutral information, is subject to top-down control. The results also have implications for models of emotional regulation (Ochsner, 2005
). We suggest that regions including the dorsolateral prefrontal cortex and parietal regions may contribute to emotional regulation by manipulating the salience of task-relevant stimuli at the expense of emotional stimuli. This function could operate in parallel to the regulatory impact that medial prefrontal cortex is thought to have on the amygdala (Quirk & Gehlert, 2003
; Pezawas et al., 2005
). Future work concerning the impact of emotional stimuli on goal-directed behavior will help determine the relative importance of each, and the parameters that determine its function.