Distinct frontal and amygdala correlates of change detection for facial identity and expression

Abstract

Recruitment of 'top-down' frontal attentional mechanisms is held to support detection of changes in task-relevant stimuli. Fluctuations in intrinsic frontal activity have been shown to impact task performance more generally. Meanwhile, the amygdala has been implicated in 'bottom-up' attentional capture by threat. Here, 22 adult human participants took part in a functional magnetic resonance change detection study aimed at investigating the correlates of successful (vs failed) detection of changes in facial identity vs expression. For identity changes, we expected prefrontal recruitment to differentiate 'hit' from 'miss' trials, in line with previous reports. Meanwhile, we postulated that a different mechanism would support detection of emotionally salient changes. Specifically, elevated amygdala activation was predicted to be associated with successful detection of threat-related changes in expression, over-riding the influence of fluctuations in top-down attention. Our findings revealed that fusiform activity tracked change detection across conditions. Ventrolateral prefrontal cortical activity was uniquely linked to detection of changes in identity not expression, and amygdala activity to detection of changes from neutral to fearful expressions. These results are consistent with distinct mechanisms supporting detection of changes in face identity vs expression, the former potentially reflecting top-down attention, the latter bottom-up attentional capture by stimulus emotional salience.

Keywords: amygdala; change detection; expression; faces; identity; prefrontal.

Fig. 1.

Example face trial. On each trial, two pictures of houses or faces were presented for 250 ms either side of a fixation cross. After a 1000 ms gap, with only the fixation cross remaining, a second image pair of the same stimulus class was presented for 250 ms. Participants’ task, upon subsequent presentation of a question mark, was to indicate by key-press whether there had been a change in the stimulus on the left (right index finger), in the stimulus on the right (right middle finger) or in neither stimulus (right ring finger). The trial shown here is a large ‘neutral to fear’ expression change trial where the face on the right remains the same pre and post interval. The face on the left retains the same identity but changes in expression from neutral to predominantly fearful (40% neutral, 60% fearful). The other conditions of primary interest comprised small expression changes (neutral to 60% neutral, 40% fearful) and small and large identity changes (see Materials and Methods).

Fig. 2.

Correlations between residual performance scores, after regressing out performance on house change trials. Participants performed consistently across small and large change trials within each type of change (ID = identity or Exp = expression) (A) but showed far less consistency of performance within each size of change, across change type(B). This suggests the presence of distinct factors influencing identity vs expression change detection performance, across participants. (See also Table 1 for corresponding partial correlation analyses. Note d.f. for correlating residuals are 1 higher than for the partial correlations.)

Fig. 3.

Regional activation for successfully detected vs missed changes on facial identity (ID) and facial expression (Exp) change trials. (A) Activity in both left and right FFA (ROIs shown in upper section of panel) was significantly greater for hits compared to misses for both identity and expression change trials. (B) VLPFC activity was greater for hits than misses for identity change trials but not for expression change trials. (C) Left amygdala activity was greater for hits vs misses for large expression changes (trials where expression changed from neutral to 60% fearful). Across participants, the magnitude of left amygdala activity associated with this contrast was positively correlated with performance on these large expression change trials (controlling for general change detection ability as indexed by performance on house change trials). *P < 0.05, **P < 0.005. Beta values are in arbitrary units.