Meta-analysis of functional brain imaging in specific phobia


Correspondence: Jonathan C. Ipser, PhD, Department of Psychiatry and Mental Health, University of Cape Town, E36A, Groote Schuur Hospital, Anzio Road, Cape Town 7925, South Africa. Email:


Although specific phobia is a prevalent anxiety disorder, evidence regarding its underlying functional neuroanatomy is inconsistent. A meta-analysis was undertaken to identify brain regions that were consistently responsive to phobic stimuli, and to characterize changes in brain activation following cognitive behavioral therapy (CBT). We searched the PubMed, SCOPUS and PsycINFO databases to identify positron emission tomography and functional magnetic resonance imaging studies comparing brain activation in specific phobia patients and healthy controls. Two raters independently extracted study data from all the eligible studies, and pooled coordinates from these studies using activation likelihood estimation, a quantitative meta-analytic technique. Resulting statistical parametric maps were compared between patients and healthy controls, in response to phobic versus fear-evoking stimuli, and before and after therapy. Thirteen studies were included, comprising 327 participants. Regions that were consistently activated in response to phobic stimuli included the left insula, amygdala, and globus pallidus. Compared to healthy controls, phobic subjects had increased activation in response to phobic stimuli in the left amygdala/globus pallidus, left insula, right thalamus (pulvinar), and cerebellum. Following exposure-based therapy widespread deactivation was observed in the right frontal cortex, limbic cortex, basal ganglia and cerebellum, with increased activation detected in the thalamus. Exposure to phobia-specific stimuli elicits brain activation that is consistent with current understandings of the neuroanatomy of fear conditioning and extinction. There is evidence that the effects of CBT in specific phobia may be mediated through the same underlying neurocircuitry.

PATIENTS WITH SPECIFIC phobia react with excessive fear to and conscious avoidance of phobic stimuli, such as small animals, heights, and blood and injury. The person must be aware that this fear is excessive and the resulting anxiety response may take the form of a situationally bound or situationally predisposed panic attack.[1] Specific phobia is a prevalent psychiatric disorder, with lifetime prevalence as high as 12.5%.[2] Although many of those with specific phobia do not present for treatment, it is a remarkably impairing condition.[3]

Functional neuroimaging has made a significant contribution to our understanding of the neuroanatomical correlates of anxiety disorders, with several functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) studies focusing specifically on specific phobia.[4] Data from these studies are crucial in attempts to understand the nature of specific phobia, as well as the biological basis of response to treatment. For instance, one model of the etiology of specific phobia proposes that phobias emerge as a result of fear conditioning processes, following prior exposure to the feared stimuli.[5] Consistent evidence for functional pathology of structures such as the amygdala and insula in specific phobia support this model, given the emphasis in recent animal models of fear conditioning of the role of the amygdala in responding to external threat and of the insula in responding to internal visceral stimuli.[6, 7]

The interpretation of the neural correlates of phobic responses in individuals diagnosed with specific phobias has unfortunately been hampered by inconsistent data from functional neuroimaging studies. This is evident in the case of the amygdala, in which diagnosis-specific activation has been reported for some studies,[8-10] but not others.[11, 12] Likewise, similar inconsistencies in the research literature are observed with respect to activation of the insula,[8, 9, 11, 12] thalamus[13] and hippocampus[12] in specific phobia. These discrepancies may to some extent be due to differences in the design and implementation of fMRI paradigms, scanning parameters and study sample across studies. For instance, there is evidence that findings may vary as a function of whether studies used event-related versus block designs.[8] This is particularly pertinent with regards to the amygdala, in that it has been hypothesized that block-designed tasks do not have the temporal resolution to control for differences between brain hemispheres in the speed at which this structure activates and subsequently habituates.[14, 15] Other factors that may explain inconsistencies across studies include differences in signal strength (e.g. 1.5 vs 3 T) and the type of specific phobia diagnosed (e.g. small animal phobia vs blood injection injury phobia).

This paper represents an attempt to address these inconsistencies in the literature through use of quantitative data synthesis techniques to expose consistent neural correlates of the phobic response. In addition, we hoped to identify changes in pathological patterns of brain activation following exposure-based therapy. In order to enhance the clinical validity of the findings, we restricted inclusion of data to functional neuroimaging studies of participants diagnosed with specific phobia according to established diagnostic criteria. The meta-analytic approach used in this study is not only less susceptible to the introduction of subjective bias than traditional narrative reviews, but also could help identify the degree to which methodological or clinical differences between studies account for the variation in study results. For instance, questions surrounding the importance of specific phobia subtype, as well as stimulus type (phobic stimuli vs generally fear-evoking stimuli) in symptom provocation paradigms could potentially be addressed using this methodology.

The current meta-analysis extends previous quantitative reviews of functional neuroimaging in specific phobia, by providing a more up-to-data synthesis of data from a larger sample of studies.[4] It is also the only imaging meta-analysis to date to address the effects of treatment in this patient type.[16] It was predicted on the basis of existing literature that greater activation would be observed in the phobic group relative to the controls in the amygdala and insula, and that this pattern of hyperactivation would dissipate following treatment.


Only studies in which participants were diagnosed with specific phobia according to DSM-III-R or DSM-IV criteria and that conducted whole-brain analyses were eligible for inclusion in the review. Studies without a healthy control group were included, provided that the subjects with specific phobia were exposed to both phobic and reference stimuli, and could therefore serve as their own controls. Studies including participants with comorbid psychiatric diagnoses were not included in this review.

We identified eligible studies by searching the PubMed, SCOPUS and psycINFO databases in January 2012 (contact the corresponding author for the full search queries used), using combinations of database-specific index and free-text terms to identify studies of subjects diagnosed with specific phobia (‘specific phobia’, ‘simple phobia’, ‘phobia’) who were scanned using at least one of a range of imaging modalities (‘functional magnetic resonance imaging’, ‘positron emission tomography’, ‘single photon emission computed tomography’, ‘computer assisted tomography’).

Two raters independently extracted data from all eligible studies, with stereotactic (x, y, z) coordinate data aggregated as part of an activation likelihood estimation (ALE) meta-analysis. ALE involves the generation of a statistical parametric map (SPM) of brain activity through the quantitative synthesis of whole-brain coordinate data across multiple studies. The likelihood that activation in particular voxels occurs by chance can subsequently be determined through reference to an empirically derived probabilistic map of brain activity.[17, 18] ALE makes optimal use of the voxel-wise resolution of the study-level data, overcomes between-study heterogeneity in the positioning of activated voxels introduced through measurement error, and bypasses reliance on subjective and error-prone anatomical labeling.[17] Moreover, conclusions drawn from applications of this method can be statistically defended through reference to a null hypothesis distribution.

Study-level data, including stereotactic coordinates, were entered independently by two raters into an electronic spreadsheet designed in accordance with the Brainmap database format. The best-fitting icbm2tal algorithm[19] was used to transform coordinates from MNI to Talairach space prior to the analysis. The false discovery rate (FDR) method was applied as a correction for multiple comparisons, given its ability to maximize sensitivity while minimizing false positive results.[20] All ALE SPM were thresholded at P < 0.05 and used a minimum cluster size of 160 mm2.

ALE SPM were generated by the GingerALE (version 2.1.1) utility provided by the authors of the Brainmap database.[21] GingerALE 2.1.1 synthesizes the location coordinate data using a random effects model, which adjusts for within-study variance through weighing a study's contribution towards the probability that a particular voxel is activated. This is achieved by adjusting the smoothing (Gaussian) kernel's full-width half maximum (FWHM) parameter downwards for larger studies, on the assumption that larger studies provide more reliable estimates of activation. The use of a random-effects model allows one to generalize the findings of the meta-analysis beyond those studies that are included within the particular analysis. GingerALE2 also restricts voxels of interest to those areas of the brain that have a >10% probability of containing gray matter.

Separate ALE maps were created for within-subject comparisons of activations in response to phobic stimuli, relative to neutral stimuli, for patients diagnosed with specific phobia and normal controls. The effect of diagnosis on neural activation in response to phobic stimuli was subsequently determined by conducting a between-group ALE analysis, with activation specific to the phobic response isolated by comparing SPM generated in the clinical group for phobic versus fearful stimuli. Finally, GingerALE 1.2 was used to conduct a separate fixed-effects ALE analysis of data from treatment trials, to determine the efficacy of cognitive behavioral therapy (CBT) in normalizing brain activity on exposure to phobic stimuli in patients diagnosed with specific phobia. A fixed-effects analysis was chosen because it has greater power than a random-effects analysis, all else being equal; an important consideration given the small samples used in this study. The comparison involved conducting separate ALE analyses (FWHM = 15 mm) on coordinate data reported before and after treatment with CBT.[12, 22, 23] Treatment effects were subsequently isolated through subtracting the post-treatment from the pre-treatment ALE maps.

The robustness of the results obtained from the treatment comparison to variability between studies was determined through repeating this analysis within a random effects framework using GingerALE 2.1.1 (with parameters as specified in the previous section). Subtraction-based hypothesis testing in GingerALE 2.1.1 is conducted with reference to a null distribution of probability estimates derived from a bootstrap estimation procedure, in which randomly selected experiments from the pool of all experiments included in the analysis are used to generate two groups of experiments of the same size as those included in the review. The SPM generated from these groups are subsequently subtracted from one another to yield ALE difference maps, with the difference statistics averaged across 10000 iterations to obtain the null distribution ALE SPM used to determine the likelihood of the foci observed in the review.[24]

Although Britton et al.,[25] Martis et al.,[26] and Schienle et al.[27] were eligible for inclusion in the review, coordinates from those studies were not included in the meta-analyses. Martis et al. was the only eligible study assessing neuropsychological performance without an affective component (implicit sequence learning), whereas no differences were observed for either Britton et al. or Schienle et al. in whole-brain between-group comparisons of subjects.


Description of studies

The search strategy identified 88 potentially eligible studies. Of these, 16 reported on independent data sets and satisfied all inclusion criteria (Fig. 1).

Figure 1.

Flowchart of study selection process. fMRI, functional magnetic resonance imaging.

Of the 16 eligible studies, 13 studies provided coordinate data, and could be included in the meta-analyses (Table 1). A total of 327 participants were included in this review, with an average age ranging between 21 and 37 years. Most of the participants were female (95%). The majority of the studies (n = 11) included individuals with spider phobia, with smaller numbers testing individuals classified as having non-spider small animal phobias (n = 3) and blood injection injury phobia (n = 1). Ahs et al. and Carlsson et al. compared 16 phobic participants by source of phobia, with eight subjects diagnosed with spider phobia and the remainder with snake phobia.[9, 28] With the exception of three PET studies,[9, 28, 31] all of the included studies used fMRI. All participants were diagnosed according to DSM-IV criteria, with the exception of the subjects in Rauch et al.,[31] who were diagnosed according to DSM-III-R criteria.

Table 1. Studies included in review
Primary authorYearModalitySample% femaleStimulusDesignSpecificationsROINo. coordinates
  1. Total number of coordinates included in all comparisons, with total number of coordinates extracted from paper in parentheses. Studies included in treatment comparison. ACC, anterior cingulate cortex; dACC, dorsal anterior and anterior mid-cingulate cortex; DLPFC, dorsolateral prefrontal cortex; DMPFC, dorsomedial prefrontal cortex; EPI, echo planar imaging; fMRI, functional magnetic resonance imaging; MPFC, medial prefrontal cortex; OFC, orbital prefrontal cortex; PAG, periaqueductal gray; PET, positron emission tomography; ROI, region of interest; SMA, supplementary motor area; TE, echo time; TR, repetition time; VLPFC, ventrolateral prefrontal cortex; VMPFC, ventromedial prefrontal cortex.
Ahs[28]2009PET8 spider and 8 snake phobics100%ImagesBlock1.5 T, 10 MBq/kg bodyweight of [15O]-water, smoothing: 12 mm5 (15)
Carlsson[9]2004PET8 spider and 8 snake phobics100%ImagesBlockI.v. bolus injections of [15O] butanolAmygdala, OFC, anterior insula, ACC, PAG (brainstem)24 (26)
Dilger[8]2003fMRI10 spider phobia and 10 healthy controls100%ImagesEvent-related1.5 T, T2*, EPI, TE = 60ms, TR = 2.3 s, smoothing: 6 mmAmygdala7 (26)
Goossens[13]2007fMRI15 spider phobics and 14 healthy controls86%ImagesEvent-related3T, T2*, EPI, TE = 30ms, TR = 2 s, smoothing: 10 mmAmygdala, thalamus19 (39)
Goossens[22]2007fMRI20 spider phobics and 14 healthy controls94%ImagesEvent-related3T, T2*, EPI, TE = 30ms, TR = 2 s, smoothing: 10 mmAmygdala, ACC, insula214 (260)
Hermann[29]2007fMRI9 BII phobics and 10 healthy controls100%ImagesBlock1.5 T, T2*, EPI, TE = 60ms, TR = 3 s, smoothing = 9 mmAmygdala, ACC, DLPFC, DMPFC, hippocampus, insula, lateral OFC, SMA, inferior and superior parietal cortex, thalamus, VMPFC10 (31)
Hermann[30]2009fMRI16 spider phobics100%ImagesEvent-related1.5 T, T2*, EPI, TE = 50ms, TR = 2860ms, smoothing = 8 mmAmygdala, Insula, rostral ACC, dACC, DLPFC, DMPFC, VLPFC, VMPFC13 (105)
Paquette[12]2003fMRI12 spider phobics and 13 healthy controls100%ImagesBlock1.5 T, T2*, EPI, TR = 0.8ms, TE = 44ms,Prefrontal cortex, visual associative cortex21 (21)
Rauch[31]1995PET7 small animal phobics86%Imagining objectsBlockTracer: 15O-CO2, smoothing: 20 mmAmygdaloid complex, anterior temporal cortex, ACC, parahippocampal gyrus, medial OFC, thalamus6 (6)
Schienle[32]2005fMRI10 spider phobia and 13 healthy controls100%ImagesBlock1.5 T, T2*, EPI, TA = 100ms, TE = 60ms, smoothing: 9 mmAmygdala, insula, lateral OFC, DLPFC, fusiform gyrus, hippocampus, MPFC16 (53)
Schienle[33]2007fMRI28 spider phobics (treatment and waiting groups) and 25 healthy controls100%ImagesBlock1.5 T, T2*, EPI, TE = 60ms, smoothing: 9 mmAmygdala, ACC, insula, medial/lateral OFC, DLPFC, fusiform gyrus, parahippocampal gyrus36 (49)
Straube[23]2006fMRI28 spider phobia (treatment and waiting groups) and 14 healthy controls100%ImagesBlock1.5 T, T2*, EPI, TE = 50ms, TR = 3.9ms, smoothing: 8 mmDLPFC, anterior insula, ACC, amygdala, parahippocampal gyrus, thalamus, fusiform gyrus82 (158)
Schweckendiek[5]2011fMRI15 spider phobics and 14 healthy controls86%ImagesEvent-related1.5 T, T2* EPI, TE = 55ms, TR = 2.5 s, smoothing = 9 mmBilateral amygdala, ACC, MPFC, bilateral OFC, bilateral thalamus and bilateral insula15 (124)

Treatment with CBT was reported in three of the eligible studies.[12, 22, 23] All of the treatment studies used a before–after experimental design, in which group therapy was carried out between scans, and in which individuals were de-sensitized through repeated exposures, over single[22] or multiple sessions,[12, 23] to the object of their phobia, at increasing levels of intensity. Treatment was regarded as successful for all of the patients who participated in these studies.

Findings of imaging meta-analysis

Table 2 lists the number of studies and coordinates upon which the findings of the meta-analytic comparisons reported in this review are based.

Table 2. No. foci (experiments) contributing to group comparisons
ComparisonNo. foci (studies)
Patients, phobic214 (11)
Controls, phobic53 (7)
Patients vs controls, phobia45 (7)
Patients, phobic > fearful 83 (6)
 Before treatmentAfter treatment
  1. †Foci that fell outside of Talairach mask in GingerALE are not included.
Patients, after vs before treatment157 (3)64 (3)

Areas of increased activation in individuals with specific phobia in response to phobic stimuli include the bilateral insula and amygdala, as well as the right medial and superior frontal cortex and the extrastriate visual cortex (Table 3). The largest cluster of activation was observed in the left insula. Most of the activation for the control group, in contrast, was observed in the visual association and higher order sensory association areas (including regions of the temporal and parietal cortex).

Table 3. Activation clusters in phobics and controls in response to phobic stimuli
 RegionHemisphereBAVolume (mm3)Weighted center (x,y,z)Extrema valuexyz§
  1. Coordinates for the center point of the cluster. Most extreme t statistic for the voxels in the cluster (×102). §Coordinates for the voxel with the largest t statistic. BA, Brodmann area.
Specific phobic group
2Medial frontal gyrusR617283.8612.9540.33.65241244
3AmygdalaL 312−20.87−3.4−9.842.077−22−4−10
4AmygdalaR 27221.060.32−15.042.19220−14
5InsulaR 23238.6913.66−0.672.11838140
6Middle occipital gyrusR1820028.25−81.97−4.422.28728−82−4
7Superior frontal gyrusR619210.53−1.9365.842.17210−266
1Inferior occipital gyrusR1868024.86−84.59−11.841.32526−86−12
2Inferior temporal gyrusR 55244.48−68.311.121.56344−680
3Inferior occipital gyrusL19368−44.22−73.53−1.211.447−44−74−2
4Anterior cingulateL323520.4735.1115.621.50503616
5Middle temporal gyrusL39344−48.89−65.6125.251.481−48−6626
6Superior parietal lobeL7256−29−49471.304−29−4947
7Parietal lobe/postcentral gyrusR319254.35−15.0740.641.156254−1640
8Inferior parietal lobuleR4016046.6−56.6934.11.13846−5634

In the direct comparison of regions of the brain that were significantly more active among the individuals with specific phobia than control subjects in response to phobic stimuli, the largest cluster contained the left amygdala and globus pallidus, followed by the right thalamus (pulvinar), left insula, and right cerebellum (Table 4; Fig. 2). Removal of the data for the single study that included BII subjects[29] did not result in significant changes to the pattern of activation observed.

Figure 2.

Activation in patients and controls in response to phobic stimuli. Clusters of brain activation in (red) social phobic patients and (blue) healthy controls.

Table 4. Greater activation in patients than controls in response to phobic stimuli
 RegionHemisphereBAVolume (mm3)Weighted center (x,y,z)Extrema valuexyz§
  1. Coordinates for the center point of the cluster. Most extreme t statistic for the voxels in the cluster (×102). §Coordinates for the voxel with the largest t statistic. BA, Brodmann area.
1Globus pallidus and amygdalaL 1048−21.39−1.76−8.361.756−180−8
2Thalamus, pulvinarR 4961.12−293.661.5412−284
4Cerebellum, decliveR 26434.01−61.11−13.081.34334−62−14

Greater limbic and striatal involvement was observed when comparing the neural response of specific phobia patients to phobic versus generally fear-evoking stimuli (Table 5). The largest clusters were observed in the left claustrum and bilaterally in the cingulate gyrus. In addition, phobia-specific activation was also observed in the right occipital gyrus and left globus pallidus.

Table 5. Greater activation in patients towards phobic vs generally fear-evoking stimuli
 RegionHemisphereBAVolume (mm3)Weighted center (x,y,z)Extrema valuexyz§
  1. Coordinates for the center point of the cluster. Most extreme t statistic for the voxels in the cluster (×102). §Coordinates for the voxel with the largest t statistic. BA, Brodmann area.
1ClaustrumL 1072−31.3316.333.293.024−32164
2Cingulate gyrusR and L3210403.619.3229.823.20761830
3Occipital, lingual gyrusR1875221.35−95.01−3.292.76422−94−4
4Lentiform nucleus, globus pallidusL 224−10.931.232.452.081−1022

Treatment with CBT was associated with a diffuse pattern of deactivation in the individuals with specific phobia in fixed-effects analysis (Table 6; Fig. 3). Large clusters (>10000 mm3) of de-activation were observed in the right frontal (superior gyrus) and bilaterally in the limbic cortex (cingulate gyrus), as well as the left insula and cerebellum. The left thalamus, right inferior frontal gyrus, right cerebellum and basal ganglia (substantia nigra) were also well represented. Smaller clusters of de-activation (<1000 mm2) included the right inferior and middle frontal cortex and left parietal cortex. No differences in brain activation following treatment were observed in the random-effects analysis.

Figure 3.

(red) Change in brain activation following treatment with cognitive behavioral therapy. (green) Regions activated in response to phobic stimuli in patients. Clusters overlaid on template in Talairach space (

Table 6. Change in brain activation following treatment with CBT
 RegionHemisphereBAVolume (mm3)Weighted center (x,y,z)Extrema valuexyz§
  1. Coordinates for the center point of the cluster. Most extreme t statistic for the voxels in the cluster (×103). §Coordinates for the voxel with the largest t statistic. BA, Brodmann area; CBT, cognitive behavioral therapy.
 1Frontal and limbic, superior frontal gyrus and anterior cingulateR6162160.8615.339.22−5.5610668064
 2Sub-lobar, insulaL1312280−39.58.772.72−1.2746269−4442
 3Posterior and cerebellum, declive and tuberL 12200−33.31−64.4−28.13−4.7374237−18−88−20
 4Thalamus, medial dorsal nucleusL 5296−0.57−−2−128
 5Cerebellum, culmen.R 296840.13−55.93−24.78−6.37645440−56−24
 6Brainstem, substantia nigraR 11047.12−11.41−8.6−5.59481148−12−8
 7Parietal, supramarginal gyrusL40976−56.62−40.1136.84−4.5399372−56−4036
 8Frontal, inferior frontal gyrusR4688848.2940.259.67−4.764645448408
 9Frontal, middle frontal gyrusL10552−30.8339.820.62−4.420075−304020
10Frontal, inferior frontal gyrusR4453647.827.1415.92−4.803895548816


The main findings of this study were that the largest clusters of activation in patients with specific phobia in response to phobic stimuli were observed bilaterally in the anterior insula and amygdala and in the right frontal cortex, and that of these regions, activation in the left insula and amygdala (in a cluster extending into the basal ganglia) remained when the phobic subjects were compared to control subjects. In addition, the right thalamus and cerebellum were activated to a greater extent in the phobic subjects than controls. Activation that was specific to the phobic relative fear response was most apparent in the left claustrum and bilaterally in the anterior cingulate and in the right occipital cortex and left pallidus. Finally, following treatment of specific phobia with CBT there was diffuse de-activation bilaterally in the anterior cingulate gyrus and in the left insula and cerebellum, with deactivation of the right frontal cortex and basal ganglia also observed. The only region to demonstrate increased activation following CBT was in the left thalamus.

The observation that the amygdala and insula are responsive to phobic stimuli in specific phobia (both when assessed in the patient group on its own and in comparison to healthy controls) is consistent with basic research on the fear response, as well as with clinical imaging data.[4] These regions overlap with activation observed in fear-conditioning paradigms,[34] supporting the model of phobias as originating from learnt fear responses.[4] Further support for the relevance of the fear-conditioning model in explaining specific phobias comes from spatial overlap between the activation pattern observed in this review in the clinical sample when exposed to phobic stimuli, and the blood oxygenation level-dependent (BOLD) response reported by Schweckendiek et al. among subjects with spider phobias relative to controls when exposed to geometric figures that had been paired with unconditioned phobic stimuli (right anterior cingulate cortex and insula, bilateral amygdala and thalamus).[5]

The largest activation in response to phobic stimuli in the patients was observed in the left insula, with corresponding de-activation of this region following treatment (Fig. 3). Debates surrounding the involvement of the insula in anxiety disorders emphasize its role in disgust sensitivity, interoception and integration of various sensory data.[7] It appears to be instrumental in discriminating between sensory information in terms of its emotional content, with some evidence that its specific association in this review with the fearful component of the phobic response may generalize to individuals who are hyper-responsive to a range of emotions.[35] The amygdala has traditionally been characterized as being more exclusively responsive to aversive stimuli, although recent literature has also placed emphasis on a more general role for this structure as being crucial for the detection of the salience or ‘arousability’ of emotional stimuli regardless of valence.[36] In addition, increased activation of the right thalamus (pulvinar) is consistent with findings that right amygdala activation to fearful faces is mediated by the pulvinar and superior colliculus.[37] This supports the involvement of a fast subcortical pathway to amygdala activation in the fear response that may not rely, at least initially, on conscious appraisal of the phobic stimulus.[6]

Activation of the anterior cingulate cortex is evident in the phobic response, and attenuates after therapy. The anterior cingulate, along with the globus pallidus (the only region where hyperactivity was both specific to the clinical group and phobic stimuli) form one of several frontal–subcortical neuronal circuits that mediate both motor and behavioral activity in humans.[38] In particular, the anterior cingulate circuit regulates motivated behavior. The cingulate is also part of an attentional network encompassing the lateral prefrontal cortex (Brodmann areas 46 and 9), parietal cortex (Brodmann area 7) and premotor and supplementary motor areas,[39] and has outflow to autonomic, visceromotor and endocrine systems. These connections facilitate the role of the anterior cingulate in assessing the salience of emotional and motivational information as well as allowing for regulation of the subsequent emotional response to these stimuli. The globus pallidum may play a role in the experience of disgust in phobic responses, and has been shown to be activated in proportion to inter-individual differences in disgust sensitivity.[40]

The largest cluster of activation found in the comparison of phobic versus fearful responses was detected in the claustrum, a structure that has reciprocal connections with many cortical regions, as well as with the hippocampus, amygdala and thalamus. Although the precise function of the claustrum is not clear, it has been hypothesized as potentially involved in the coordination of various cortical areas by integrating these inputs into conscious perception.[41] Its activation may therefore represent neural processing of fearful stimuli over a longer time-course than that which takes place in the rapid ventral subcortical pathway to the amygdala. Although the greater activation of brain regions in phobic subjects in response to phobic versus fearful stimuli suggests that it would be premature to characterize phobias as emerging solely from pathological fear-conditioning processes, these findings may at least partly reflect differences in the properties of the stimuli used in the phobic and fearful conditions.

Cerebellum activation was prominent in the patient group relative to controls when exposed to phobic stimuli, and a reduction of activation was observed in the cerebellum following therapy. The cerebellum plays a role in coordination of movement, autonomic response, and storage of fear-related memories.[42] While the basolateral complex of the amygdala has long been implicated as the primary storage site of memories involved in fear conditioning, cerebellar involvement in the long-term storage and expression of fear memories acquired by fear conditioning has also been demonstrated.[43] The present finding demonstrating involvement of the cerebellum in specific phobia may therefore relate to fear-conditioning models of phobia acquisition and expression.

The only region in which increased activation was observed after treatment was the thalamus. Previous evidence of thalamic involvement in specific phobia has been interpreted in terms of the hypothesized role of this structure in relaying signals to the amygdala as part of a subcortical circuit that allows rapid responses to environmental threats.[13] The present finding that exposure-based treatments result in increased activation of the thalamus without a corresponding response from the amygdala, suggests that CBT may be acting via the disruption of this subcortical fear-processing circuit.

Several limitations to this review should be noted. Despite the systematic use of rigorous study identification and data analytic procedures, the present results were based on relatively few studies, limiting the power to reliably detect phobia-specific brain activation. As a result, the conclusions drawn from this review cannot be regarded as definitive. In addition, the majority of the participants included in this meta-analysis were female, and it is not clear to what extent one can generalize the results to men. The finding of substantial lateralization of activation in response to phobic stimuli to the left hemisphere may partially reflect the predominance of women in the included studies, with evidence for a greater extent of activation in the left amygdala for women than men when exposed to aversive stimuli.[44, 45] Finally, the comparisons were limited to assessments of increases in brain activation in subjects with phobia, and did not test for neural hypoactivation in this group, despite some evidence of diffuse patterns of hypoactivation in phobic subjects when exposed to phobic stimuli relative to non-phobic fearful stimuli.[28]

With regard to the treatment meta-analysis, the fixed-effects analysis used is susceptible to the disproportionate influence of small studies that capitalize on chance by using liberal statistical thresholds to adjust for low power. This is a particular concern given how few studies contributed to this particular analysis, and the fact that a random-effect analysis was not able to replicate the results of the treatment comparison. Differences in the variability of the data contributed by the studies included in the treatment meta-analysis were also not accounted for, making generalizations regarding the efficacy of CBT in specific phobia subjects outside of these study contexts problematic.

The influence of phobia subtype on brain activity could not be systematically assessed, because only one study of phobias other than small animal phobia was included. Previous research suggests that blood injection injury phobia is differentiated from small animal phobia by greater sensitivity to disgust as opposed to fearful stimuli.[46] Findings from a recent series of studies indicate that there may be a differential patterning of the BOLD response when comparing blood injection injury phobia with small animal phobia, with a corresponding shift of activation from the limbic regions to the orbitofrontal cortex and visual regions.[15, 47, 48] Nevertheless, there was little evidence that inclusion of the single study of BII subjects in this review had a substantial affect on the results of the meta-analysis.

Finally, it should be noted that inclusion in this review was restricted to studies with subjects who satisfied the full set of DSM diagnostic criteria, thus excluding a number of studies that either did not require that subjects fulfill specific criteria,[47, 48] or that used cut-offs on scales of symptom severity to define specific phobia.[15, 36, 49] Although we feel that this decision was warranted, given the intent of this review to characterize disturbances in brain activity associated with the diagnosis of specific phobia, it should be acknowledged that the results of this review may not be generalizable to all individuals who evidence substantial phobic responses. Because the majority of studies assessing the neural correlates of phobic responses in BII subjects failed to diagnose specific phobia according to the full set of DSM criteria, relaxing the study inclusion criteria in future reviews would also help to characterize this subgroup of individuals.

Despite these limitations, the present results were largely in accordance with current understandings of the neuroanatomy of fear conditioning and extinction. Several further lines of investigation should be pursued, including work focused on the role of neuroreceptors relevant to phobia in these circuits, possible gender effects on brain responses to phobic stimuli, and the effects of pharmacotherapy and psychotherapy in normalizing brain activation in phobic patients. Implementations of factorial study designs that simultaneously compare neural response across groups (phobic vs healthy controls) and conditions (phobic vs fearful) would also shed light on the extent to which functional neuropathology in specific phobia differs qualitatively as well as quantitatively from brain activation in normative fear responses.


We would like to thank Dr Ines Meyer for assistance in assessing studies for eligibility, and with data extraction. Drs Vincent Paquette and Liesbet Goossens kindly provided coordinate data for the treatment arms of their studies. We are also grateful to Drs Thomas Straube and Gustav Wik for providing information on the methodological aspects of their studies.

Leesha Singh and Jonathan Ipser declare no conflicts of interest. Dan Stein has received research grants and/or consultancy honoraria from AstraZeneca, Eli-Lilly, GlaxoSmithKline, Lundbeck, Orion, Pfizer, Pharmacia, Roche, Servier, Solvay, Sumitomo, and Wyeth.