predictors of amygdala activation during the processing

8/8/2019 Predictors of Amygdala Activation During the Processing

http://slidepdf.com/reader/full/predictors-of-amygdala-activation-during-the-processing 1/14

Review

Predictors of amygdala activation during the processing of

emotional stimuli: A meta-analysis of 385 PET

and fMRI studies

Sergi G. Costafreda⁎ , Michael J. Brammer, Anthony S. David, Cynthia H.Y. Fu

Institute of Psychiatry, King's College London, UK

A R T I C L E I N F O A B S T R A C T

Article history:

Accepted 27 October 2007

Available online 12 November 2007

Although amygdala activity has been purported to be modulated by affective and non-

affective factors, considerable controversy remains on its precise functional nature. We

conducted a meta-analysis of 385 functional neuroimaging studies of emotional processing,

examining the effects of experimental characteristics on the probability of detecting

amygdala activity. All emotional stimuli were associated with higher probability of

amygdala activity than neutral stimuli. Comparable effects were observed for most

negative and positive emotions, however there was a higher probability of activation for

fear and disgust relative to happiness. The level of attentional processing affected amygdala

activity, as passive processing was associated with a higher probability of activation thanactive task instructions. Gustatory-olfactory and visual stimulus modalities increased the

probability of activation relative to internal stimuli. Aversive learning increased the

probability of amygdala activation as well. There was some evidence of hemispheric

specialization with a relative left-lateralization for stimuli containing language and a

relative right-lateralization for masked stimuli. Methodological variables, such as type of

analysis and magnet strength, were also independent predictors of amygdala activation.

© 2007 Elsevier B.V. All rights reserved.

Keywords:

Amygdala

Emotion

Neuroimaging

PET

fMRI

Meta-analysis

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.1. Results of the systematic literature review. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 592.2. Terminology and odds ratio estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.3. Emotions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.4. Study instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.5. Stimulus characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.6. Effect of other predictors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.7. Laterality interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.1. Emotions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

B R A I N R E S E A R C H R E V I E W S 5 8 ( 2 0 0 8 ) 5 7 – 7 0

⁎ Corresponding author. Institute of Psychiatry, PO Box 89, De Crespigny Park, London SE5 8AF, UK. Fax: +44 203 328 2116.E-mail address: [email protected] (S.G. Costafreda).

0165-0173/$ – see front matter © 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.brainresrev.2007.10.012

a v a i l a b l e a t w w w . s c i e n c e d i r e c t . c o m

w w w . e l s e v i e r . c o m / l o c a t e / b r a i n r e s r e v

mailto:[email protected]

http://dx.doi.org/10.1016/j.brainresrev.2007.10.012

http://dx.doi.org/10.1016/j.brainresrev.2007.10.012

mailto:[email protected]



3.2. Task instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.3. Stimulus modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.4. Social emotions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.5. Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.6. Lateralization effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.7. Other variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.8. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665. Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.1. Literature search and inclusion criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.2. Data extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.3. Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Appendix A. Supplement data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

1. Introduction

Since the classical observations characterized in the Klüver-

Bucy syndrome,it hasbeenwidely accepted that the amygdala

has a key role in many facets of emotional processing

(Aggleton and Saunders, 2000). The advent of neuroimaging

has allowed in vivo, albeit indirect, recordings of amygdala

activity in humans which has led to a dramatic increase in the

available data. However, there remains much controversy

about the precise nature of its function.

Many factors have been proposed to modulate amygdala

activity in the processing of emotional stimuli. With regards to

affective valence, the amygdala has a well-established role in

fear processing and aversive conditioning (LeDoux, 2000a,b;

Schafe and LeDoux, 2004; Adolphs et al., 1999; Davis and

Whalen, 2001), but recent formulations have emphasized its

recruitment by positive emotions (Holland and Gallagher,

2004; Baxter and Murray, 2002; Rolls, 2000a). It is known that

amygdala lesions have deleterious consequences on primate

social behavior (Kling et al., 1970; Amaral, 2002), and accord-

ingly, some authors have argued that signals of a social nature

may particularly engage the human amygdala (Barton and

Aggleton, 2000; Adolphs, 2003a,b). The sensory modality of the

stimulus presentation also affects the responsiveness of the

amygdala, with a claim for dominance by visual stimuli

(Markowitsch 1998; Barton and Aggleton, 2000). Another factor

purportedto affect its activityis hemispheric asymmetry of the

functions of the amygdala (Markowitsch, 1998). It has been

suggested that the right amygdala subserves a system of

automatic detection of emotional stimuli (Morris et al., 1998,

1999; Wright et al., 2001; Glascherand Adolphs,2003), whilethe

left amygdala is more responsive to conscious, language

dependent processing (Markowitsch, 1998; Morris et al., 1998;

Funayama et al., 2001; Phelps, 2006).

However, the empirical support for these formulations has

often been based on a relatively small number of functional

neuroimaging experiments. Questions about the generaliz-

ability of these assertions arise when evidence can be found

for both sides of a claim, for example whether emotional

stimuli elicit a reflexive, automatic activation of the amygdala

(Morris et al., 1998) or that its engagement requires conscious

attentional processing (Pessoa and Ungerleider, 2004).

These concerns are amplified when one considers that

functional neuroimaging studies usually suffer from relatively

small samples. Moreover, there are particular difficulties for

imaging the amygdala region with functional magnetic

resonance imaging (fMRI) on account of its size and magnetic

resonance field inhomogeneities related to air-tissue inter-

faces (Glover and Law, 2001; LaBar et al., 2001). Negative

findings in a particular study may be due to low power to

detect amygdala activity, rather than as a result of a true

absence of activity. The integration of data across studies is

further complicated by methodological differences, such as

the method of analysis or fMRI magnet strength, which could

independently affect the probability of amygdala activation

and thus confound any judgments.

For these reasons, the rationale for meta-analytic techni-

ques as a tool for bringing together evidence from multiple

studies is clear. An approach used in previous meta-analyses

on amygdala function has been to calculate the proportion of

studies with (and without) a feature that is associated with its

activation (Phan et al., 2002; Baas et al., 2004). Although the

“vote-counting ” method may yield valuable insights, the

appraisal is confounded by correlations between potential

predictors. For example, a recent meta-analysis reported that

two factors relevant for amygdala activation were emotions of

a fearful valence and the presentation of stimuli in the visual

modality (Phan et al., 2002). However, 8 of the 9 pertinent

studies in the meta-analysis had delivered the fearful emotion

through visual presentations. With the “vote-counting ”

approach, it is not possible to make a definite attribution of

the amygdala activation effect to either of these variables.

In the present study, we used a “meta-regression” analysis

in order to delineate the impact of multiple, concurrent

experimental factors on the probability of detecting amygdala

activation (Whitehead, 2003). In this approach, the probability

of a given experiment to detect amygdala activity is modelled

as an additive function of the experimental characteristics.

This method allows demarcation of the different effects and

the computation of independent estimates for each factor or

predictor. The data for our analysis were obtained from a

systematic literature review of the neural correlates of

emotional stimuli processing in healthy adults. We examined

the following predictors: experimental and baseline task

58 B R A I N R E S E A R C H R E V I E W S 5 8 ( 2 0 0 8 ) 5 7 – 7 0



features, and the neuroimaging acquisition and analysis

variables. In the experimental task features, we examined

affective valence, modality of presentation, presence of

language in the stimulus, the pairing with an aversive

stimulus (aversive conditioning), whether the stimulus had

been masked to prevent conscious awareness, and the study

instructions given to the participants (e.g. explicit labelling of

emotions or passive viewing). Methodological factors whichcould have an independent effect on the probability of

detecting amygdala activation, namely acquisition system:

Positron Emission Tomography (PET) or fMRI, and the type of

analysis: whole brain or region-of-interest (ROI), were also

included in our model (Costafreda et al., 2006). In addition, we

explored the possibility of hemispheric specialization of

amygdala function by examining laterality interactions for

each factor. We gave each study a weight proportional to its

sample size, an important procedure as smaller experiments

are likely to be underpowered due to the amygdala size and

anatomical location (LaBar et al., 2001).

2. Results

2.1. Results of the systematic literature review

The systematic review produced 385 studies fulfilling our

criteria, containing results on 1324 experiments from 5307

individual subjects. Full study characteristics, univariateactivation tables, and the list of references are presented as

supplementary material (Supplementary Tables 1 and 2,

Supplementary ReferenceList). Briefly, the studies were recent

as over 2/3 of the studies had been published in the past

5 years, andtherewere on average 13.8 subjects per study. The

neuroimaging method employed by the majority of the

studies (298) was fMRI. Bilateral amygdala activation was

reported in 220 experiments, unilateral right amygdala

activation in 124, and unilateral left amygdala activation in

153 experiments. Chi square tests revealed significant bivari-

ate associations ( pb0.05) between all the potential predictors

Fig. 1 – Location of the amygdala activation foci and areas of statistically significant density of activation foci. Axial (z= -12) and

coronal (y=-4) slices, in Talairach space, representing the peaks of activation for the 94 studies that reported amygdala

activation and also provided spatial coordinates (each focus is represented by a red X). The green Region-of-Interest boxes

correspond to the extreme coordinates of amygdala tissue as described in the Talairach and Tournoux Atlas ( x coordinates

from +/-14 to +/- 30, y coordinates from - 11 to + 1 and z coordinates from - 22 to - 5). Ninety-six percent of the total of activation

foci (179 foci) were contained in the ROI boxes, showing that to a large extent, the reported coordinates were consistent with the

amygdala labels. The slices on the right show the areas with statistically significant density of peaks of activation (false

discovery rate significance level q =0.05). As expected, this analysis revealed significant bilateral density, with maxima x=+/-22,

y=-6, z=-12.

59B R A I N R E S E A R C H R E V I E W S 5 8 ( 2 0 0 8 ) 5 7 – 7 0



and amygdala activation. We have also plotted the amygdala

foci for the subset of studies reporting peak activation coor-

dinates in addition to the anatomical label (Fig. 1). Activation

likelihood analysis (Laird et al., 2005a) of these foci revealed, as

expected, significant bilateral density of activations (maxima

at x=+/-22, y= -6, z= -12; Fig. 1).

2.2. Terminology and odds ratio estimates

The estimates for each factor are presented in Figs. 2–5. For

each predictor, a baseline level was determined and its

estimate set to an odds ratio (OR) value of 1 (dot on red dashed

line). An OR greater than1 signifiesthat the predictorincreases

the odds of activation relativeto the baseline level,whilean OR

smaller than 1 indicates a reduction in the odds of activation.

The horizontal segments represent the 95% confidence inter-

val (CI) for the effect estimate based on a normal approxima-

tion. If theconfidence interval does notcrossthe OR =1 vertical

line, it is statistically significant at α=0.05. The values of the

parameters are described in full in Supplementary Table 3.

2.3. Emotions

Emotional stimuli were more likely to result in amygdala

activation than neutral stimuli (Fig. 2). This increase was

significant for all emotions, with the exception of unspecified

positive emotions (OR =1.46, 95% CI= [0.73 - 2.89], p =0.28). The

magnitude of this increase varied according to emotion, but

most differences were not statistically significant as indicated

by the overlapping confidence intervals. The only exceptions

were the estimates for fear (OR =6.80, 95% CI=[4.09-11.32],

pb0.001) and disgust (OR=6.24, 95% CI=[3.36-11.58], pb0.001),

which were both significantly greater than the effect for

happiness (OR =1.88, 95% CI=[1.07-3.30], pb0.001) and unspe-

cified positive emotions. Although fear and disgust were the

most likely emotions to result in amygdala activation, they

were closely followed by humour (OR =6.22, pb0.001). The

strength of the effect of disgust may appear surprising as theunivariate results suggested that fear was a much stronger

predictor of amygdala activity: 65% of experiments using fear

resulted in amygdala activation, while the corresponding

figure for disgust was 40% (Supplementary Table 2). However,

further analysis revealed a striking dependence between the

effect of disgust and the study sample size with amygdala

activation increasing from 7% (1/15) of the experiments with

less than 10 subjects to 41% (14/34) in experiments with 10 to

20 subjects and reaching 89% (8/9) for larger (20+ subjects)

studies.

2.4. Study instructions

Task instructions involving a form of attentional processing

reduced the odds of amygdala activation compared to the

passive processing of emotional stimuli (Fig. 2). This reduction

was statistically significant for all active instructions except

for attended incidental processing of stimuli, such as perform-

ing a gender decision on emotional faces, which however

showed a trend towards significance (OR=0.67, p =0.08). The

reduction in amygdala activity was most marked when

subjects were asked to suppress their emotional response to

Fig. 2 – Effect of the type of emotion and the task instructions, on the odds of amygdala activation.




the stimuli (OR =0.07, pb0.001). Emotional suppression

resulted in significantly less amygdala activation than explicit

emotional labelling and attended incidental processing as

indicated by non-overlapping confidence intervals.

2.5. Stimulus characteristics

External stimuli were generally more likely to activate the

amygdala than internal ones (Fig. 3). This difference was

statistically significant for face stimuli (OR =4.32, pb0.01) and

visual stimuli in general (OR =3.26, pb0.01). Gustatory and

olfactory stimuli were the strongest predictor of amygdala

activation (OR =6.17, pb0.01). As well, emotional learning,

which was a form of aversive conditioning in the studies

included in the meta-analysis, was associated with a sig-

nificant increase in the odds of activation (OR =2.53, pb0.05).

2.6. Effect of other predictors

Other methodological factors were also strong predictors of

amygdala activation (Fig. 4). Contrasts with low-level control

tasks (OR =2.46, pb0.001) were more likely to detect amygdala

activation than contrasts with a neutral control condition or

those involving other emotions (OR=0.43, pb0.001). A ROI

analysis was a strong predictor of amygdala activation

(OR =3.70, pb0.001) compared to a whole-brain analysis

approach. Studies using fMRI were more likely than PET

studies to detect amygdala activation. Furthermore, in fMRI

studies, there was a positive relationship between the

increase in odds and the strength of the magnet (OR =1.62,

p =0.08 for 1.5 T; OR =1.88, p =0.17 for 2 T; OR =2.52, pb0.01 for

3 T; and OR =4.48, pb0.05 for 4 T scanners).

2.7. Laterality interactions

Laterality interactions were statistically significant with only

two factors: presenceof language in the stimulus andmasking

of the stimulus (Fig. 5). Language was associated with an

overall reduction in activation (OR =0.49, pb0.05) and a left

lateralization effect (OR =0.48 for right amygdala activation,

pb0.05). The main effect of masked stimuli was not signifi-

cant, butthere wasa strong right lateralization effect(OR=4.03

for right amygdala activation, pb0.01).

3. Discussion

3.1. Emotions

The present study modelled the probability of amygdala acti-

vation during the processing of emotional stimuli as a function

of multiple experimental characteristics. The main finding

Fig. 3 – Effect of stimulus modality and whether the stimulus has been associated with an aversive unconditioned stimulus

(learning), on the odds of amygdala activation.




indicated that all emotions were more likely to activate theamygdala than neutral stimuli. Among the emotions, there

was evidence of significant differences between subtypes in

their capacity to generate amygdala activation. Fear anddisgust were more likely to be associated with amygdala

activation than happiness and unspecified positive emotions.

Fig. 5 – Effect of the presence of language in the stimulus and whether the stimulus had been masked, on amygdala activation

and hemispheric specialization.

Fig. 4 – Effect of the type of baseline task, the type of analysis and the acquisition system, on the odds of amygdala activation.




These results may be interpreted as evidence that the

amygdala is predominantly activated by negative, rather than

positive stimuli (Paradiso et al., 1999; Ohman and Mineka,

2001). Indeed, based on the behavioral effects of selective

amygdala lesions on primates, it has been hypothesized that

the amygdala functions as a danger detection device (Amaral,

2002). Some neuropsychological data support this distinction

in humans as studies in patients with amygdala lesions havegenerally reported impairments in fear recognition and other

negative emotions, but not for positive ones (Broks et al., 1998;

Adolphs et al., 1999; Anderson and Phelps, 2000). Electro-

physiological recordings in patients with refractory epilepsy

have also noted significant changes with negative but not

positive pictures (Oya et al., 2002).

However, the specificity of the link between amygdala

lesion and fear recognition impairment in humans has

recently been questioned by both large-scale patient studies

(Rapcsak et al., 2000) and re-examination of the deficits of

individual patients with bilateral amygdala lesions (Adolphs

et al., 2005). Fear is the most difficult emotion to recognize, not

only for amygdala lesion patients but also for subjects with

other brain lesions and even healthy subjects (Rapcsak et al.,

2000). In addition, a potential mechanism for the impairments

in amygdala patients could be a general difficulty in the use of

information from the region of the eyes in analysing emo-

tional faces of any valence, which may be critical for fear

recognition but not for other emotions (Adolphs et al., 2005).

Indeed, there is an extensive experimental literature indicat-

ing that the role of the amygdala is not limited to negative

emotions. In humans, single neuron recordings have found

that theamygdala codes for the recognition of emotional facial

expressions in general, not selectively for fearful faces (Fried

et al., 1997). In rodents, the amygdala has been demonstrated

to play a role in emotional learning of positive associations

(reviewed in Holland and Gallagher, 2004; Baxter and Murray,

2002) and in sexual behavior (Zald, 2003). Electrophysiological

recordings in primates have shown that the amygdala

responds to a wide range of emotions (Rolls, 2000a), from

unconditioned positive stimuli such as food (Nishijo et al.,

1988) to emotional happy faces (Nakamura et al., 1992).

Gothard and colleagues (2007) have recently studied cells in

the monkey amygala selectively responding to positive,

neutral and negative facial expressions. Although the number

of emotionally-selectivecellswas similar for the three types of

emotions, there was a small but significant increase on the

overall firing rate for negative stimuli relative to the other two

types. As discussed by the authors, such a differential effect

would result in subtraction-based functional neuroimaging

studies finding a net increase of activation for negative

emotions. These observations are therefore largely consistent

with our results.

In our meta-analysis, positive emotions such as humour

and sex were both strong predictors of amygdala activation. Of

note, humour was the thirdstrongest emotion associated with

amygdala activation. If happiness and humour may be con-

sidered to be separated less by differences in valence than by

differences in arousal, our findings may be viewed as indirect

evidence that arousal drives amygdala activation, rather than

valence (Anderson and Sobel, 2003; Small et al., 2003). The

relative differences in effect sizes between emotions would

then be the result of the differences in the arousal capacity of

the stimuli, not from a specific effect of emotional valence.

The strength of the effect of sexual stimuli in predicting

amygdala activation supports this interpretation, as partici-

pants routinely rated these stimuli as highly arousing. How-

ever, a difficulty in equating amygdala function with arousal

evaluation arises with anger which is an emotion usually

considered as highly arousing (Posner et al., 2005), but it wasassociated with only a modest increase on amygdala activa-

tion, particularly in comparison with fear. An alternative

proposal to account for this apparent anomaly has been

stimulus ambiguity as there is high ambiguity in the potential

threat source for fear but low uncertainty for anger (Davis and

Whalen, 2001).

The strong effect of disgust may seem surprising as many

neuroimaging studies have reported amygdala activation with

fear, but not disgust, in the same sample of subjects (for

example: Phillips et al., 1997), which has led to the supposition

that the amygdala is involved in fear but not disgust pro-

cessing (Calder et al., 2001). The results of our meta-analysis

suggest that the effect of disgust was large, but there was a

stronger dependence on study sample size than with other

emotions. A possible explanation may be that the amygdala

effects generated by disgust are more variable across subjects

than those with other emotions, and greater variability across

individuals would necessitate larger sample sizes to demon-

strate a significant effect. An intriguing notion is whether the

variability in amygdala activity could be related to the con-

siderable inter-subject variability in the sensitivity to disgust

as disgust sensitivity has been correlated with the strength of

amygdala activity (Schienle et al., 2005).

3.2. Task instructions

Passive processing was associated with a higher probability of

amygdala activation than tasks involving any form of atten-

tional effort, namely explicit and implicit emotional proces-

sing. Moreover, the strongest difference was found between

passive attentional processing and active emotional suppres-

sion instructions. An increase in cognitive demands by active

attentional processing may lead to greater amygdala inhibi-

tion. Fear conditioning, a form of learning dependent on

amygdala function (LeDoux, 2000b), is progressively disrupted

by the increasing difficulty of a concurrent memory task

(Carter et al., 2003). Theinhibition of amygdala activity may be

mediated by an increased recruitment of frontal cortical

regions. The anterior cingulate and lateral prefrontal cortices

areengaged by diverse cognitivedemands (Duncan and Owen,

2000) and deemed essential for cognitive control during

effortful tasks (Matsumoto and Tanaka, 2004). These medial

and dorsolateral frontal areas produce an inhibitory effect on

amygdala function, which has been observed in rodents

(Rosenkranz et al., 2003; Quirk et al., 2003; Sotres-Bayon et al.,

2004) and humans(Hariri et al., 2003; Pezawas et al., 2005). The

anterior cingulate cortex projects to the amygdala (Aggleton

and Saunders, 2000), its activity increases with task difficulty

(Paus et al., 1998; Fu et al., 2002; Botvinick et al., 2004), and it

shows a negative correlation with amygdala activity (Pezawas

et al., 2005). Moreover, Blairand colleagues(2007) reported that

increasing task difficulty was associated with a significant




linear decrease in bilateral amygdala activation, while activity

in the anterior cingulate and dorsolateral prefrontal cortices

were positively correlated to each other but negatively with

amygdala activity.

An adaptive functional role has been suggested for the

coupling of task difficulty and reduction in amygdala activa-

tion. Automatic responses to affective stimuli can disrupt

performance during demanding goal-directed behavior ascompared to neutral stimuli, which is exemplified by the

emotional Stroop task (Pratto and John, 1991) and observed in

other tasks with emotional distractors (Simpson et al., 2000;

Tipples and Sharma, 2000; Hartikainen et al., 2000; Blair et al.,

2007). The attentional bias towards emotional stimuli and its

consequent disruptive effect on performance is likely to be

mediated by the amygdala (Anderson and Phelps, 2001). Top-

down inhibition of amygdala activation during effortful, goal-

directed behavior could therefore be seen as a mechanism of

hierarchical integration (Serrien et al., 2006) to ensure main-

tenance of performance dependent on cortical areas in the

presence of potentially disrupting emotional stimuli.

Among the active tasks instructions, the demand to down

regulate the response to emotional stimuli resulted in the

lowest probability of amygdala activation. Studies used two

strategies: reappraisal, in which the meaning of the emotional

stimulus was manipulated to enhance the emotional experi-

ence; and suppression, in which subjects were asked to

‘distance themselves’ emotionally from the stimuli (Gross,

2002). It has been suggested that the magnitude and timing of

emotional responses may be regulated through prefrontal

inhibitory circuits (Rosenkranz et al., 2003). Such a modulatory

role is also supported by studies of emotional regulation in

humans, which have identified the dorsal anterior cingulate

and dorsolateral prefrontal cortices as important areas during

voluntary regulation (Ochsner and Gross, 2005). Moreover,

dorsal anterior cingulate activity is correlated with the

amount of reduction in negative affect generated by down-

regulation through reappraisal (Ochsner et al., 2002). This

suggests that voluntary emotional regulation may act through

potentiation of pre-existing mechanisms for hierarchical

cortical inhibition of amygdala function.

Among the active processing instructions, there were few

significant differences. However, there is some evidence that

task difficulty could explain the relative differences between

active task instructions. For example, explicit labelling of

emotions is a more difficult task as compared to common

incidental tasks such as making a gender-decision about

affective facial expressions (Rapcsak et al., 2000; Kapler et al.,

2001; Hariri et al., 2003). Accordingly, we found that explicit

labellingwas associated with a relative decrease in theodds of

activation relative to incidental processing.

An alternative explanation proposes that the absence of

amygdala activity may be due to a ‘passive’ relative lack of

attentional resources in the presence of competing processing

demands, rather than from an ‘active’ top-down inhibition.

Pessoa and Ungerleider (2004) reported that in the presence of

a highly demanding, non-emotional task, the presentation of

emotional faces was not associated with amygdala activation.

In the absence of the distracting task, they did observe amyg-

dala activation suggesting that the amygdala response to

emotional stimuli is dependent on attention.

3.3. Stimulus modality

The effect of gustatory and olfactory stimuli was greater than

for other sensory modalities. The amygdala receives both

direct and indirect gustatory connections, which have been

linked to motivational aspects of taste (Spector, 2000).

Olfactory input has a uniquely direct access to the amygdala,

and it has been argued that some of the amygdala nuclei maybe considered as the main olfactory association cortex

(Swanson, 2000). Visual stimuli were associated with a higher

probability of amygdala activation as compared to internal

and auditory stimuli, although this difference was not

significant relative to auditory stimuli. It has been suggested

that the amygdala may be specialized for the processing of

visual stimuli (Markowitsch, 1998), which may be related to

the intense connectivity and coevolution of the amygdala and

the visual ventral stream across primate species (Barton and

Aggleton, 2000). Internal emotional stimuli, which were often

generated by recollection of autobiographical episodes or with

script-driven imagery, were associated with a relatively low

probability of amygdala activation compared to other mod-

alities. Internal stimuli depend to some extent on mental

imagery capabilities (Kosslyn et al., 2001), which show con-

siderable interindividual differences (Isaac and Marks, 1994).

These interindividual differences could result in less reliable

activations across individuals to internal stimuli and there-

fore, to a lower power of functional studies to detect sig-

nificant differences compared to external modalities.

3.4. Social emotions

The amygdala has been identified as an important area for

primate social behavior (Kling et al., 1970; Amaral, 2002). Our

findings add to a complementary body of evidence on the

importance of social signals to human amygdala function

(reviewed in Adolphs, 2003b). Not only was the probability of

amygdala activation for social emotions significantly above

that for neutral stimuli, but emotional faces which are an

inherently social type of stimuli were strong predictors of

amygdala activation. It has been suggested that facial visual

stimuli have a primary, unlearned reinforcing value (Slater

et al., 1998; Rolls, 2000a), and accordingly in our meta-analysis,

faces resulted in a moderately higher probability of amygdala

activation than other visual stimuli.

3.5. Learning

Aversive learning was also a strong predictor of amygdala

activation. The dependence of aversive learning on amygdala

function is well established in both animal andhumanstudies

(for reviews: Phelps and LeDoux, 2005; LaBar and Cabeza,

2006). Amygdala damage impairs aversive conditioning (LaBar

et al., 1995; Bechara et al., 1995), and its activity is predictive of

the strength of the conditioned response (Phelps et al., 2004;

LaBar et al., 1998).

3.6. Lateralization effects

The presence of language in the stimulus was associated with

a lower probability of detecting amygdala activation. In many




experiments, the stimuli were written language. Understand-

ing language is an effortful act, and the reduction in prob-

ability of amygdala activation could be linked to its inhibition

by prefrontal regions (Rosenkranz et al., 2003; Pezawas et al.,

2005; Blair et al., 2007). As well, neuropsychological data

indicate that emotional knowledge encoded in language may

be processed in the absence of amygdala (Bechara et al., 1995).

There was also a significant laterality interaction for lan-guage as the reduction in the probability of activation was

stronger in the right hemisphere, resulting in a significant

relative left-lateralization. Phelps and colleagues (2001)

demonstrated that emotional information conveyed exclu-

sively through language (i.e. the threat of an electric shock)

resulted in unilateral left amygdala activation. In temporal

lobectomy patients, those with left amygdala lesions failed

to show potentiation of the startle response following a ver-

bally communicated threat, while both control subjects and

right amygdala patients showed a typical startle response

(Funayama et al., 2001). The normal perceptual bias for

aversive relative to neutral words is absent in patients with

bilateral or left amygdala lesions, but not with right amygdala

damage (Anderson and Phelps, 2001). Preferential activation of

the left amygdala by language could be part of the general left

hemispheric dominance for language (Schirmer and Kotz,

2006), and some authors further suggest a specialization for

language function in the left amygdala (Phelps et al., 2001;

Markowitsch, 1998).

Stimuli that were manipulated to prevent awareness, most

frequently through temporal masking, resulted in the opposite

pattern with a relative lateralization to the right amygdala.

These findings are in accordance with the model in which the

right amygdala mediates an initial, fast and perhaps automatic

stimulus detection role, followed by a more discriminative and

evaluative response that is subserved by the left amygdala

(Morris et al., 1998; Wright et al., 2001; Glascher and Adolphs,

2003; Adolphs, ). Thisfastresponsemay bemediated bya direct,

subcortical thalamic pathway through a colliculo-pulvinar-

amygdala route (LeDoux et al., 1984; Morris et al., 1999),

responsive to the coarse features of the stimulus (Vuilleumier

et al., 2003), while the delayed, evaluative response is depen-

dent on an indirect, cortical route to the amygdala.

In support, patients with right amygdala lesions show

significantly lower skin conductance responses to backward

masked emotional pictures than both healthy controls and

patients with left amygdala lesions (Glascher and Adolphs,

2003). A patient with cortical blindsight correctly identified

sad, fearful and happy faces in association with right

amygdala activation with an accuracy that was above chance

(Pegna et al., 2005). However, evidence for a right-lateralized,

rapid detection system by the amygdala has been limited to

emotional facial expressions as the blindsight patient had

poor discrimination for the emotional valence of complex

scenes. As well, all the experiments included in our meta-

analysis had used emotional faces as subliminal stimuli.

Whether such a rapid emotional detection system is exclusive

for facial stimuli or would be activated by other evolutionarily

valuable and coarsely recognisable stimuli (e.g. snakes)

requires further study.

This differential hemispheric lateralization for language

and masked stimuli mayalso provide a neural substrateto the

observations by Olsson and Phelps (2004). In their experiment

an association was made between a conditioned stimulus

(angry face) and either a real shock, the verbal threat of a

shock, or the observation of a shock given to someone else.

When the angry face was overtly presented, all three groups

showed evidence of conditioning as measured by increased

skin conductance relative to an unconditioned angry face.

When the stimulus was masked, only the group with thelanguage dependent link failed to show the same evidence of

conditioning. Our results reinforce the author's conclusion

that the lack of response may be due to a relative failure of

right lateralized information about the masked stimulus

reaching left lateralized language knowledge of its learned

value.

3.7. Other variables

The type of baseline control condition was a significant pre-

dictor of amygdala activation. As expected, low level baseline

tasks, such as fixation or rest, were associated with a higher

probability of activation than control conditions involving

emotional or neutral stimuli.

Task independent variables of type of analysis and acqui-

sition system had strong and significant effects that were

similarin magnitudeto theeffects of emotion subtype. Whether

the analysis was based on a amygdala ROI or a whole-brain

approach had a strong effect on the probability of amygdala

detection. It is likely that this effect originates from the less

stringent multiple comparisons correction required for an ROI

approach. This effect also suggests that whole-brain analysis is

a relatively insensitive procedure for the detection of amygdala

activation.

PET studies were less likely to detect amygdala activation

perhaps due to lower spatial and temporal resolution, partic-

ularly in older systems. In addition, the progressiveincreasein

odds ratio at higher fields for fMRI seems to indicate that their

theoretical advantages are matched by empirical improve-

ments. The magnitude of these improvements was similar to

the ones generally described in the literature (Friedman et al.,

2006). Importantly, the effects of ‘methodological’ predictors

(for a discussion, see Zald,2003) highlightthe need foraccurate

reporting of methodological parameters and the potential for

confounded conclusions if these factors are not taken into

account (Costafreda et al., 2006; Murphy et al., 2003; Paus et al.,

1998) in a meta-analysis or even a qualitative review.

3.8. Limitations

Our meta-analysis was limited by the quality of information

retrievable from published functional neuroimaging papers,

which were often incomplete in quantifiable and retrievable

data. Typically, only the anatomical label or peak coordinates

of areas deemed significantly active at the group level were

reported. Group level data generally obscures a great deal of

complexity, such as gender related interactions (Cahill et al.,

2004). Our capabilities for meta-analysis would be substan-

tially enhanced if subject-level functional data, which are

likely to encapsulate a very large part of the variability

(Costafreda et al., 2007), were provided. As well, current

limitations in spatial resolution prevent delineation of




amygdala nuclei, which are likely to have specific functions

(Phelps and LeDoux, 2005; Everitt et al., 2003). Despite high

intraamygdalar connectivity (Pitkanen, 2000) and correlated

volumetric evolution of amygdala subnuclei across species

(Barton et al., 2003), there is reason to be cautious in

presuming a unified set of functions for the whole structure

(Swanson and Petrovich, 1998).

Some of the conclusions in our study are based on a limitednumber of experiments (Supplementary Table 2). In our view,

this is an inescapable consequence of the methodological

heterogeneity between functional neuroimaging studies. It is

rarely the case that a study is identically replicated. More

often, even related studies are different in important char-

acteristics: for example, two studies may have used the same

or similar tasks, but it is likely that they differ in some other

importantaspect, such as scanner systemor analysismethod.

The most common approach to this heterogeneity in the

neuroimaging meta-analysis literature has been the use of

restrictive inclusion criteria to achieve homogeneity in the

included papers, often leading to small sample sizes.

Our approach overcomes this limitation, in part, by adjust-

ing in the analysis for sources of variability between studies.

The probability of amygdala activation is modelled as a linear

function of each study characteristics. This allows a relaxation

of the inclusion criteria, but it does not free our work from the

requirement of a sufficient number of experiments necessary

so that these sources of variability can be adequately

estimated. That is, if we included predictors that have been

used by very low number of experiments, it is very doubtful

that we could provide a precise estimate of the variability

induced by such factors, thus endangering the whole model-

ling. In practice, the minimum number of experiments for a

given characteristic was 15, while all other conclusions were

based on 25 or more experiments. In our view, this not only

allows the computation of reasonably precise estimates, but

also compares favorably with the majority of previously

published meta-analysis.

In our analysis, we used a linear additive model to predict

amygdala activation which is parsimonious but is likely a

simplification of reality. As more is learned about amygdala

function,more complicatedmodels maybe built. In particular,

the additivity assumption is questionable for gustatory and

olfactory stimuli as some complex emotions may not be

conveyed through those modalities. In order to test for this

possible effect, we re-analyzed the data with the exclusion of

gustatory and olfactory experiments, but found no important

differences and all significant variables remained accordingly.

Also, we applied relatively conservative criteria in our model-

ling process. Although we did not find evidence for further

lateralization effects, our findings should not be taken as

excluding their existence.

Many other stimuli characteristics, which mayappear to be

more subtle, also affect amygdala activation, namely gaze

direction in face processing (Adams et al., 2003). Unfortu-

nately, covariates such as gaze direction could not be recorded

in our data collection due to inconsistent reporting. Conscious

of this limitation, we allowed study-specific and experiment-

specific random intercepts, which would cope to some extent

with this unaccounted variability, by allowing each study and

experiment within study their own specific baseline prob-

ability of activation. However, our task analysis was in this

sense admittedly crude, again reflecting what information can

be consistently retrieved from published studies with a

sufficient degree of certainty. It is clear though that a meta-

analysis performed with even relatively simple models and

less than ideal data can yield interesting insights, which

complement experimental evidence and when unexpected

results are found may fuel the hypothesis generation process.

4. Conclusions

Therefore,in spite of its limitations, some global suggestions on

amygdala function do emerge from our meta-analysis. Amaral

conceived the amygdala as a “protection device (…) designed to

detect and avoid danger ” (Amaral, 2002). Our findings are

consistent with an extension of this view towards a more

general role of emotional evaluation of incoming stimuli,

whereby meaning is attached to behaviorally relevant stimuli.

Negative stimuli may be privileged, perhaps due to their often

direct survival value, but positive ones also generate amygdala

activation. In accordance with this relevance detection role, we

also found that external stimuli are prioritized over internally

generated ones. This detection function may also be facilitated

by passive perception, and inhibited either by voluntary sup-

pression or by co-occurring higher-level activity such as lan-

guage. We found evidence of hemispheric specialisation, where

therightamygdalamay subserve a high-speed detectionrole for

unconscious stimuli, while the left amygdala may be preferen-

tially recruited when evaluation of language-related stimuli is

required.

5. Methods

5.1. Literature search and inclusion criteria

A systematic literature search was conducted through medical

databases(EMBASE,MEDLINE andPsycINFO)to identify PETand

fMRI journal articles published from January 1990 to July 2006

presentingoriginal results on the neuralcorrelatesof processing

emotional stimuli. Inclusion criteria were studies reporting

negative or positive results on amygdala activation resulting

from a contrast between an active and a baseline task in the

same subjects, healthy adults of 18 to 65 years of age. In clinical

studiesreporting separate resultsfrom a patient population and

a group of healthy controls, only thecontrol data were included.

Exclusion criteria were tactile sensory stimuli, lack of bilateral

amygdala coverage, experimental designs consisting of a

psychoactivedrug or placebo, and between groupscomparisons

(e.g. men vs women) as the results of these experiments are

determined not only by our predictors but also by group

differences. Studies were also detected through manual exam-

ination of reference lists of original studies and review papers

identified by the computerized search.

5.2. Data extraction

The data extraction was performed from the journal article

and publicly available supplementary material. Amygdala




activation was coded as a binary variable (0= not active,

1=active), and the hemispheric laterality was recorded. Both

region of interest, label-based and coordinate-based activa-

tions were included, employing the authors' terminology. The

number of subjects in each study was recorded. We extracted

the following experimental characteristics as model predic-

tors, which may modulate amygdala activation:

• Experimental task features: emotion subtype, experimental

task instructions, modality of presentation of the stimuli,

use of language in the emotional stimuli, masking of the

stimuli, whether the stimuli had been conditioned through

aversive learning, and the baseline, control task.

• Methodological variables: acquisition system (PET or fMRI),

magnet strength for fMRI, type of analysis.

Emotions can be subdivided according to a number of

categorizations (Adolphs, 2002). In our study, we included the

five basic emotions (fear, disgust, anger, sadness, happiness),

along with humour and sex, two other positive emotions of

interest because they have recently been used in a number of

experiments. We also included the category of social emo-

tions, representing emotions that are dependent on a social

context and which participate in the regulation of social

behavior (Adolphs, 2003a). Social emotions were thus guilt,

embarrassment, shame, abandonment, pride, admiration,

attachment, friendship, love and moral dilemmas (Adolphs,

2003a). For studies which did not clearly fulfil the previous

emotion categories or had stimuli representing more than one

of the emotion subtypes, we employed the following cate-

gories: positive emotion not specified (e.g. pleasantness,

beauty), negative emotion not specified (e.g. combinations of

negative emotions: fear and disgust), and emotion not

specified which encompassed experiments combining posi-

tive and negative emotional valences in the same stimulus or

emotions that were not easily categorized as positive or

negative (e.g. surprise).

Experimental task instructions were classified according to

the following categories: passive experience of the emotion;

explicit processing, in which the task required an explicit

judgement of the emotional stimulus; and incidental proces-

sing, when the task instruction involved a non-emotional

task. Explicit processing experiments were further classified

into three categories: labelling: explicit labelling or rating of

the emotional stimulus; feeling: feeling the emotion elicited

by the stimulus; and emotional suppression: reduction of the

emotion elicited by the stimulus. Incidental processing tasks

were also subdivided into two categories: attended: in which

the attention of the subject was focused on non-emotional

features of the emotional stimuli (e.g. gender decision for

faces with affective expressions); and distracted: when a non-

emotional task was used to shift attention from the emotional

study (e.g. an oddball task presented simultaneously with

emotional stimuli).

The modality of the stimuli presentation was coded as:

visual, auditory, audio-visual, gustatory/olfactory (GO), inter-

nal (i.e. memory recall and imagery), and a combined internal

and external category for studies that used both internal and

external probes (e.g. recalling a negative autobiographical

memory while watching a sad face). Visual stimuli of “Ekman”

or “Ekman-like” faces (Ekman and Friesen, 1976) were

recorded separately. The presence of faces in more complex

visual stimuli (e.g. IAPS pictures, video clips) could not be

reliably ascertained and was therefore not recorded. Any

unequivocal use of language in the emotional stimuli,

whether visual, auditory or internal (e.g. in the recall of

emotional sentences), was also modelled as an individual

variable. In our coding protocol, we did not try to imputewhether the internal processing mechanisms of the experi-

mental stimuli were dependent on language, but only whether

the stimulus explicitly contained language. For example, if the

subject had to imagine a sad event, the experiment was not

coded as containing language as the task could have been

accomplished by visual imagery alone. The use of a procedure

to suppress conscious awareness of the experimental stimu-

lus was also coded as a separate predictor. The pairing of the

experimental stimulus to an unconditioned stimulus was also

recorded as a variable. In the present review, the uncondi-

tioned stimuli employed in the experiments were all aversive.

The control task was classified in three categories based on

the type of stimuli employed: emotional, neutral, and low

level (e.g. a resting state control condition).

The type of acquisition system was coded as PET or fMRI

and, for fMRI, magnet strength was recorded. Analysis was

classified as region of interest (ROI) if stated in the study or if

a less stringent threshold was applied in the amygdala than

in other brain areas, such as a small volume correction of

only the amygdala volume. Details of the tasks, stimuli and

sample size of each study are presented in Supplementary

Table 1.

5.3. Statistical analysis

The unadjusted association between amygdala activation and

its potential predictors was assessed using contingency tables

and Pearson's chi-square tests, with Yates'continuity correc-

tion when appropriate. For the subset of studies reporting the

coordinates of amygdala activation, an activation likelihood

estimation analysis (Laird et al., 2005a), with 5000 permutation

and a false discovery rate significance level q= 0.05, was

performed to reveal areas of significant density of activation

foci using the BrainMap tools (www.brainmap.org ; Laird et al.,

2005b) for conversion of coordinates to the Talairach space

(Talairach and Tournoux, 1998), creation of ROI “boxes” and

plotting. A mixed-effects logistic regression model (Whitehead,

2003) was fitted tothe data toestimatethe adjusted effect of the

predictors, that is the effect of a given predictor independently

of any of the other variables in the model. The fixed effects in

the model were the following: laterality, type of emotion,

stimulus modality, stimulus contains language or not, stimulus

masked or not, type of study instructions, type of control task,

analysis and acquisition system. The random effects consisted

of random intercepts for study and experiment within study,

assumed to be independent of each other and of the residual

variability. This formulation allowed each study and experi-

ments within eachstudy their own specific baseline probability

of activation, acknowledging the numerous unmeasured

covariates that were likely to affect the probability of amygdala

activation but could not be recorded in our data collection

related to inconsistent reporting in the original studies. These


http://www.brainmap.org/

http://www.brainmap.org/



random intercepts also modelled the degree of correlation

introduced in our data by the hierarchical structure of experi-

ments within studies. Ignoring this hierarchical structure and

theinduced correlation(i.e. fittinga model with thefixedeffects

only) would generate smaller standard errors for the effect

estimates and thus over-significant results (PinheiroandBates,

2000).

To investigate differential lateralization effects, we used astepwise model selection strategy. The minimum model we

were willing to consider included only the main effects of the

predictors, while the maximum model included the main

effects and all the pairwise first-level interaction of laterality

and predictors. The stepwise procedure included or excluded

variables based on the Akaike Information Criteria (AIC,

Akaike, 1974) of the fitted model. The model with the lowest

AIC waschosen, reflecting the best compromise between good

model fit and simplicity of the model. As the AIC criteria can

sometimes be too liberal, we checked the significance of the

two included interactions using ANOVA F-tests, confirming

that both interactions were statistically significant (language

by amygdala laterality and masked stimuli by laterality

interactions, both pb0.001). Because of marked methodologi-

cal differences between PET and fMRI, notably in spatial

resolution, it may be doubted whether studies using both

techniques can be combined, or should be reported separately.

In our analysis, a significant interaction between acquisition

technique and other substantive variables, such as type of

emotion, would suggest that PET and fMRI provide very

different experimental results, and therefore combining

studies across imaging techniques is potentially misleading.

We investigated the existence of such interactions following

the same steps as for laterality interaction, and found no

evidence of a significant interaction between acquisition

technique and other predictors.

All the models were fitted using Penalized Quasi-Like-

lihood (Breslow and Clayton, 1993), employing the number

of subjects per experiment as a prior weight. The standard

packages MASS (Venables and Ripley, 2002) and nlme (Pin-

heiro and Bates, 2000) of the R statistical software

(R Development Core Team, 2005, http://www.R-project.org )

were employed for the analysis. The number of subjects in

each study wasused to weighthe contributionof each studyto

parameters estimates. A P-value of less than 5% was

considered to indicate statistical significance, and all tests

were two-sided. Confidence intervals were obtained by using a

normal approximation based on the estimated standard errors

of the parameters, with a confidence level of 0.95. In the text

we have used the term amygdala ‘activation’, as termed in the

literature, whenever an experiment declared that there was

evidence for an increase in neural activity in the amygdala as

measured through statistically significant local haemody-

namic changes of blood flow (for PET studies) or blood oxygen

level dependent (BOLD) response (for fMRI studies) with the

active task.

Acknowledgments

SGC is supported by a Medical Research Council (UK) Fellow-

ship in Neuroinformatics.

Appendix A. Supplementary data

Supplementary data associated with this article can be found,

in the online version, at doi:10.1016/j.brainresrev.2007.10.012.

R E F E R E N C E S

Adams, R.B.J., Gordon, H.L., Baird, A.A., Ambady, N., Kleck, R.E.,2003. Effects of gaze on amygdala sensitivity to anger and fear faces. Science 300, 1536.

Adolphs, R., 2002. Neural systems for recognizing emotion. Curr.Opin. Neurobiol. 12, 169–177.

Adolphs, R., 2003a. Cognitive neuroscience of human socialbehavior. Nat. Rev., Neurosci. 4, 165–178.

Adolphs, R., 2003b. Is the human amygdala specialized for processing social information? Ann. N. Y. Acad. Sci. 985,326–340.

Adolphs, R., Tranel, D., Hamann, S., Young, A.W., Calder, A.J.,Phelps, E.A., Anderson, A., Lee, G.P., Damasio, A.R., 1999.

Recognition of facial emotion in nine individuals with bilateralamygdala damage. Neuropsychologia 37, 1111–1117.

Adolphs, R., Gosselin, F., Buchanan, T.W., Tranel, D., Schyns, P.,Damasio, A.R., 2005. A mechanism for impaired fear recognition after amygdala damage. Nature 433, 68–72.

Aggleton, J.P., Saunders, R.C., 2000. The amygdala- what'shappened in the last decade. In: Aggleton, J.P. (Ed.), TheAmygdala: a Functional Analysis. Oxford University Press,Oxford, pp. 1–30.

Akaike, H., 1974. New look at the statistical model identification.IEEE Trans. Automat. Contr. 19, 716–723.

Amaral, D.G., 2002. The primate amygdala and the neurobiology of social behavior: implications for understanding social anxiety.Biol. Psychiatry 51, 11–17.

Anderson, A.K., Phelps, E.A., 2000. Expression without recognition:

contributions of the human amygdala to emotionalcommunication. Psychol. Sci. 11, 106–111.

Anderson, A.K., Phelps, E.A., 2001. Lesions of the human amygdalaimpair enhanced perception of emotionally salient events.Nature 411, 305–309.

Anderson, A.K., Sobel, N., 2003. Dissociating intensity fromvalence as sensory inputs to emotion. Neuron 4, 581–583.

Baas, D., Aleman, A., Kahn, R.S., 2004. Lateralization of amygdalaactivation: a systematic review of functional neuroimaging studies. Brain Res. Brain Res. Rev. 45, 96–103.

Barton, R.A., Aggleton, J.P., 2000. Primate evolution and theamygdala. In: Aggleton, J.P. (Ed.), The Amygdala: a FunctionalAnalysis. Oxford University Press, Oxford, pp. 479–508.

Barton, R.A., Aggleton, J.P., Grenyer, R., 2003. Evolutionarycoherence of the mammalian amygdala. Proc. Biol. Sci. 270,539–543.

Baxter, M.G., Murray, E.A., 2002. The amygdala and reward. Nat.Rev., Neurosci. 3, 563–573.

Bechara, A., Tranel, D., Damasio, H., Adolphs, R., Rockland, C.,Damasio, A.R., 1995. Double dissociation of conditioning anddeclarative knowledge relative to the amygdala andhippocampus in humans. Science 269, 1115–1118.

Blair, K., Smith, B., Mitchell, D., Morton, J., Vythilingam, M., Pessoa,L., Fridberg, D., Zametkin, A., Sturman, D., Nelson, E., Drevets,W., Pine, D., Martin, A., Blair, R., 2007. Modulation of emotion bycognition and cognition by emotion. NeuroImage 35, 430–440.

Botvinick, M.M., Cohen, J.D., Carter, C.S., 2004. Conflict monitoring and anterior cingulate cortex: an update. Trends Cogn. Sci. 8,539–546.

Breslow, N.E., Clayton, G.D., 1993. Approximate inference ingeneralized linear mixed models. J. Am. Stat. Assoc. 88, 9–25.


http://www.r-project.org/

http://dx.doi.org/doi:10.1016/j.brainresrev.2007.10.012

http://dx.doi.org/doi:10.1016/j.brainresrev.2007.10.012




Broks, P., Young, A.W., Maratos, E.J., Coffey, P.J., Calder, A.J., Isaac,C.L., Mayes, A.R., Hodges, J.R., Montaldi, D., Cezayirli, E.,Roberts, N., Hadley, D., 1998. Face processing impairmentsafter encephalitis: amygdala damage and recognition of fear.Neuropsychologia 36, 59–70.

Cahill, L., Uncapher, M., Kilpatrick, L., Alkire, M.T., Turner, J., 2004.Sex-related hemispheric lateralization of amygdala function inemotionally influenced memory: an fmri investigation. Learn.

Mem. 11, 261–266.Calder, A.J., Lawrence, A.D., Young, A.W., 2001. Neuropsychology

of fear and loathing. Nat. Rev., Neurosci. 2, 352–363.Carter, R.M., Hofstotter, C., Tsuchiya, N., Koch, C., 2003. Working

memory and fear conditioning. Proc. Natl. Acad. Sci. U. S. A.100, 1399–1404.

Costafreda, S.G., Fu, C.H., Lee, L., Everitt, B., Brammer, M.J., David,A.S., 2006. A systematic review and quantitative appraisal of fMRI studies of verbal fluency: Role of the left inferior frontalgyrus. Hum. Brain Mapp. 27, 799–810.

Costafreda, S.G., Brammer, M.J., Vêncio, R.Z., Mourão,M.L., Portela,L.A., de Castro, C.C., Giampietro, V.P., Amaro Jr., E., 2007.Multisite fMRI reproducibility of a motor task using identicalMR systems. J. Magn. Reson. Imaging 26, 1122–1126.

Davis, M., Whalen, P.J., 2001. The amygdala: vigilance and

emotion. Mol. Psychiatry 6, 13–34.Duncan, J., Owen, A.M., 2000. Common regions of the human

frontal lobe recruited by diverse cognitive demands. TrendsNeurosci. 23, 475–483.

Ekman, P., Friesen, V.W., 1976. Pictures of Facial Affect. Consulting Psychologists Press, Palo Alto, CA.

Everitt, B.J., Cardinal, R.N., Parkinson, J.A., Robbins, T.W., 2003.Appetitive behavior: impact of amygdala-dependentmechanisms of emotional learning. Ann. N. Y. Acad. Sci. 985,233–250.

Fried, I., MacDonald, K.A., Wilson, C.L., 1997. Single neuron activityin human hippocampus and amygdala during recognition of faces and objects. Neuron 18, 753–765.

Friedman, L., Glover, G., Krenz, D., Magnotta, V., 2006. Reducing inter-scanner variability of activation in a multicenter fMRI

study: Role of smoothness equalization. NeuroImage 32,1656–1668.

Fu, C.H.Y., Morgan, K., Suckling, J., Williams, S.C.R., Andrew, C.,Vythelingum, G.N., McGuire, P.K., 2002. A functional magneticresonance imaging study of overt letter verbal fluency using aclustered acquisition sequence: greater anterior cingulateactivation with increased task demand. NeuroImage 17,871–879.

Funayama, E.S., Grillon, C., Davis, M., Phelps, E.A., 2001. A doubledissociation in the affective modulation of startle in humans:effects of unilateral temporal lobectomy. J. Cogn. Neurosci. 13,721–729.

Glascher, J., Adolphs, R., 2003. Processing of the arousal of subliminal and supraliminal emotional stimuli by the humanamygdala. J. Neurosci. 23, 10274–10282.

Glover, G.H., Law, C.S., 2001. Spiral-in/out bold fMRI for increasedSNR and reduced susceptibility artifacts. Magn. Reson. Med. 46,515–522.

Gothard, K.M., Battaglia, F.P., Erickson, C.A., Spitler, K.M., Amaral,D.G., 2007. Neural responses to facial expressions and faceidentity in the monkey amygdala. J. Neurophysiol. 97,1671–1683.

Gross, J.J., 2002. Emotion regulation: affective, cognitive, and socialconsequences. Psychophysiology 39, 281–291.

Hariri, A.R., Mattay, V.S., Tessitore, A., Fera, F., Weinberger, D.R.,2003. Neocortical modulation of the amygdala response tofearful stimuli. Biol. Psychiatry 53, 494–501.

Hartikainen, K.M., Ogawa, K.H., Knight, R.T., 2000. Transientinterference of right hemispheric function due toautomatic emotional processing. Neuropsychologia 38,1576–1580.

Holland, P.C., Gallagher, M., 2004. Amygdala-frontal interactionsand reward expectancy. Curr. Opin. Neurobiol. 14, 148–155.

Isaac, A.R., Marks, D.F., 1994. Individual differences in mentalimagery experience: developmental changes andspecialization. Br. J. Psychol. 85, 479–500.

Kapler, E., Hariri, A., Mattay, V., McClure, R., Weinberger, D., 2001.Correlated attenuation of amygdala and autonomic responses:A simultaneousfMRIand SCRstudy. Soc. Neurosci.Abstr. 645,3.

Kling, A., Lancaster, J., Benitone, J., 1970. Amygdalectomy in thefree-ranging Vervet (Cercopithecus Aethiops). J. Psychiat. Res.,Suppl. 7, 191–199.

Kosslyn, S.M., Ganis, G., Thompson, W.L., 2001. Neuralfoundations of imagery. Nat. Rev., Neurosci. 2, 635–642.

Laird, A.R., Fox, M., Price, C.J., Glahn, D.C., Uecker, A.M., Lancaster, J.L., Turkeltaub, P.E., Kochunov, P., Fox, P.T., 2005a. ALEmeta-analysis: Controlling the false discovery rate andperforming statistical contrasts. Hum. Brain Mapp. 25, 155–164.

Laird, A.R., Lancaster, J.L., Fox, P.T., 2005b. BrainMap: The socialevolution of a functional neuroimaging database.Neuroinformatics 3, 65–78.

LaBar, K.S., Cabeza, R., 2006. Cognitive neuroscience of emotionalmemory. Nat. Rev., Neurosci. 7, 54–64.

LaBar, K.S., LeDoux, J.E., Spencer, D.D., Phelps, E.A., 1995. Impaired

fear conditioning following unilateral temporal lobectomy inhumans. J. Neurosci. 15, 6846–6855.

LaBar, K.S., Gatenby, J.C., Gore, J.C., LeDoux, J.E., Phelps, E.A., 1998.Human amygdala activation during conditioned fear acquisition and extinction: a mixed-trial fMRI study. Neuron20, 937–945.

LaBar, K.S., Gitelman, D.R., Mesulam, M.M., Parrish, T.B., 2001.Impact of signal-to-noise on functional MRI of the humanamygdala. NeuroReport 12, 3461–3464.

LeDoux, J., 2000a. The amygdala and emotion: a view through fear.In: Aggleton, J.P. (Ed.), The Amygdala: a Functional Analysis.Oxford University Press, Oxford, pp. 289–310.

LeDoux, J.E., 2000b. Emotion circuits in the brain. Annu. Rev.Neurosci. 23, 155–184.

LeDoux, J.E., Sakaguchi, A., Reis, D.J., 1984. Subcortical efferent

projections of the medial geniculate nucleus mediateemotional responses conditioned to acoustic stimuli.

J. Neurosci. 4, 683–698.Markowitsch, H., 1998. Differential contribution of right and left

amygdala to affective information processing. Behav. Neurol.11, 233–244.

Matsumoto, K., Tanaka, K., 2004. Conflict and cognitive control.Science 303, 969–970.

Morris, J.S., Ohman, A., Dolan, R.J., 1998. Conscious andunconscious emotional learning in the human amygdala.Nature 393, 467–470.

Morris, J.S., Ohman, A., Dolan, R.J., 1999. A subcortical pathway tothe right amygdala mediating “unseen” fear. Proc. Natl. Acad.Sci. U. S. A. 96, 1680–1685.

Murphy, F.C., Nimmo-Smith, I., Lawrence, A.D., 2003. Functionalneuroanatomy of emotions: a meta-analysis. Cogn. Affect.Behav. Neurosci. 3, 207–233.

Nakamura, K., Mikami, A., Kubota, K., 1992. Activity of singleneurons in the monkey amygdala during performance of avisual discrimination task. J. Neurophysiol. 67, 1447–1463.

Nishijo, H., Ono, T., Nishino, H., 1988. Single neuron responses inamygdala of alert monkey during complex sensory stimulationwith affective significance. J. Neurosci. 8, 3570–3583.

Ochsner, K.N., Gross, J.J., 2005. The cognitive control of emotion.Trends Cogn. Sci. 9, 242–249.

Ochsner, K.N., Bunge, S.A., Gross, J.J., Gabrieli, J.D.E., 2002.Rethinking feelings: an fMRI study of the cognitive regulationof emotion. J. Cogn. Neurosci. 14, 1215–1229.

Ohman, A., Mineka, S., 2001. Fears, phobias, and preparedness:toward an evolved module of fear and fear learning. Psychol.Rev. 108, 483–522.




Olsson, A., Phelps, E.A., 2004. Learned fear of “unseen" faces after pavlovian, observational, and instructed fear. Psychol. Sci. 15,822–828.

Oya, H., Kawasaki, H., Howard, M.A.R., Adolphs, R., 2002.Electrophysiological responses in the human amygdaladiscriminate emotion categories of complex visual stimuli.

J. Neurosci. 22, 9502–9512.Paradiso, S., Johnson, D.L., Andreasen, N.C., O'eary, D.S., Watkins,

G.L., Ponto, L.L., Hichwa, R.D., 1999. Cerebral blood flowchanges associated with attribution of emotional valence topleasant, unpleasant, and neutral visual stimuli in a pet studyof normal subjects. Am. J. Psychiatry 156, 1618–1629.

Paus, T., Koski, L., Caramanos, Z., Westbury, C., 1998. Regionaldifferences in the effects of task difficulty and motor output onblood flow response in the human anterior cingulate cortex: areview of 107 PET activation studies. NeuroReport 9, 37–47.

Pegna, A.J., Khateb, A., Lazeyras, F., Seghier, M.L., 2005.Discriminating emotional faces without primary visualcortices involves the right amygdala. Nat. Neurosci. 8, 24–25.

Pessoa, L., Ungerleider, L.G., 2004. Neuroimaging studies of attention and the processing of emotion-laden stimuli. Prog.Brain Res. 144, 171–182.

Pezawas, L., Meyer-Lindenberg, A., Drabant, E.M., Verchinski, B.A.,

Munoz, K.E., Kolachana, B.S., Egan, M.F., Mattay, V.S., Hariri,A.R., Weinberger, D.R., 2005. 5-HTTLPR polymorphism impactshuman cingulate-amygdala interactions: a geneticsusceptibilitymechanism for depression. Nat. Neurosci. 8, 828–834.

Phan, K.L., Wager, T., Taylor, S.F., Liberzon, I., 2002. Functionalneuroanatomy of emotion: a meta-analysis of emotionactivation studies in pet and fmri. NeuroImage 16, 331–348.

Phelps, E.A., 2006. Emotion and cognition: insights from studies of the human amygdala. Annu. Rev. Psychol. 57, 27–53.

Phelps, E.A., LeDoux, J.E., 2005. Contributions of the amygdala toemotion processing: from animal models to human behavior.Neuron 48, 175–187.

Phelps, E.A., O'Connor, K.J., Gatenby, J.C., Gore, J.C., Grillon, C.,Davis, M., 2001. Activation of the left amygdala to a cognitiverepresentation of fear. Nat. Neurosci. 4, 437–441.

Phelps, E.A., Delgado, M.R., Nearing, K.I., LeDoux, J.E., 2004.Extinction learning in humans: role of the amygdala andvmpfc. Neuron 43, 897–905.

Phillips, M.L., Young, A.W., Senior, C., Brammer, M., Andrew, C.,Calder, A.J., Bullmore, E.T., Perrett, D.I., Rowland, D., Williams,S.C., Gray, J.A., David, A.S., 1997. A specific neural substrate for perceiving facial expressions of disgust. Nature 389, 495–498.

Pinheiro, J.C., Bates, D.M., 2000. Mixed-effects Models in S andS-PLUS. Statistics and Computing. Springer-Verlag, New York.

Pitkanen, A., 2000. Connectivity of the rat amygdaloid complex. In:Aggleton, J.P. (Ed.), The Amygdala: a Functional Analysis.Oxford University Press, Oxford, pp. 31–116.

Posner, J., Russell, J.A., Peterson, B.S., 2005. The circumplex modelof affect: an integrative approach to affective neuroscience,cognitive development, and psychopathology. Dev.Psychopathol. 17, 715–734.

Pratto, F., John, O.P., 1991. Automatic vigilance: theattention-grabbing power of negative social information.

J. Pers. Soc. Psychol. 61, 380–391.Quirk, G.J., Likhtik, E., Pelletier, J.G., Pare, D., 2003. Stimulation of

medial prefrontal cortex decreases the responsiveness of central amygdala output neurons. J. Neurosci. 23, 8800–8807.

R Development Core Team, 2005m. R: a Language andEnvironment for Statistical Computing. R Foundation for Statistical Computing, Vienna. http://www.R-project.org .

Rapcsak, S.Z., Galper, S.R., Comer, J.F., Reminger, S.L., Nielsen, L.,Kaszniak, A.W., Verfaellie, M., Laguna, J.F., Labiner, D.M.,Cohen, R.A., 2000. Fear recognition deficits after focal braindamage: a cautionary note. Neurology 54, 575–581.

Rolls, E.T., 2000a. Neurophysiology and functions of the primateamygdala, and the neural basis of emotion. In: Aggleton, J.P.(Ed.), The Amygdala: a Functional Analysis. Oxford UniversityPress, Oxford, pp. 479–508.

Rosenkranz, J.A., Moore, H., Grace, A.A., 2003. Theprefrontal cortexregulates lateral amygdala neuronal plasticity and responsesto previously conditioned stimuli. J. Neurosci. 23, 11054–11064.

Schafe, G.E., LeDoux, E.J., 2004. The neural basis of fear. In:Gazzaniga, M.S. (Ed.), The Cognitive Neurosciences III. The MITPress, Cambridge, MA, pp. 987–1004.

Schienle, A., Schafer, A., Stark, R., Walter, B., Vaitl, D., 2005.Relationship between disgust sensitivity, trait anxiety andbrain activity during disgust induction. Neuropsychobiology51, 86–92.

Schirmer, A., Kotz, S.A., 2006. Beyond the right hemisphere: brainmechanisms mediating vocal emotional processing. TrendsCogn. Sci. 10, 24–30.

Serrien, D.J., Ivry, R.B., Swinnen, S.P., 2006. Dynamics of hemispheric specialization and integration in the context of

motor control. Nat. Rev., Neurosci. 7, 160–166.Simpson, J.R., Ongur, D., Akbudak, E., Conturo, T.E., Ollinger, J.M.,

Snyder, A.Z., Gusnard, D.A., Raichle, M.E., 2000. The emotionalmodulation of cognitive processing: an fMRI study. J. Cogn.Neurosci. 12, 157–170.

Slater, A., Von Der Schulenburg, C., Brown, E., Badenoch, M.,Butterworth, G., Parsons, S., Samuels, C., 1998. Newborn infantsprefer attractive faces. Infant Behav. Dev. 21, 345–354.

Small, D.M., Gregory, M.D., Mak,Y.E., Gitelman, D., Mesulam, M.M.,Parrish, T., 2003. Dissociation of neural representation of intensity and affective valuation in human gustation. Neuron39, 701–711.

Sotres-Bayon, F., Bush, D.E.A., LeDoux, J.E., 2004. Emotionalperseveration: an update on prefrontal-amygdala interactionsin fear extinction. Learn. Mem. 11, 525–535.

Spector, A.C., 2000. Linking gustatory neurobiology to behavior invertebrates. Neurosci. Biobehav. Rev. 24, 391–416.

Swanson, L.W., 2000. Cerebral hemisphere regulation of motivatedbehavior. Brain Res. 886, 113–164.

Swanson, L.W., Petrovich, G.D., 1998. What is the amygdala?Trends Neurosci. 21, 323–331.

Talairach, J., Tournoux, P., 1998. Co-planar stereotaxic atlas of thehuman brain. Thieme Medical Publishers, New York.

Tipples, J., Sharma, D., 2000. Orienting to exogenous cues andattentional bias to affective pictures reflect separate processes.Br. J. Psychol. 91, 87–97.

Venables, W.N., Ripley, B.D., 2002. Modern Applied Statistics withS. Statistics and Computing. Springer-Verlag, New York.

Vuilleumier, P., Armony, J.L., Driver, J., Dolan, R.J., 2003. Distinctspatial frequency sensitivities for processing faces andemotional expressions. Nat. Neurosci. 6, 624–631.

Whitehead, A., 2003. Dealing with heterogeneity. Meta-Analysis of Controlled Clinical Trials. Statistics in practice. Wiley,Chichester, England, pp. 151–174.

Wright, C.I., Fischer, H., Whalen, P.J., McInerney, S.C., Shin, L.M.,Rauch, S.L., 2001. Differential prefrontal cortex and amygdalahabituation to repeatedly presented emotional stimuli.NeuroReport 12, 379–383.

Zald, D.H., 2003. The human amygdala and the emotionalevaluation of sensory stimuli. Brain Res. Rev. 41, 88–123.




predictors of amygdala activation during the processing

Documents