Increased cortical thickness in a frontoparietal network in social anxiety disorder

Abstract

Social anxiety disorder (SAD) is the second leading anxiety disorder. On the functional neurobiological level, specific brain regions involved in the processing of anxiety‐laden stimuli and in emotion regulation have been shown to be hyperactive and hyper‐responsive in SAD such as amygdala, insula and orbito‐ and prefrontal cortex. On the level of brain structure, prior studies on anatomical differences in SAD resulted in mixed and partially contradictory findings. Based on previous functional and anatomical models of SAD, this study examined cortical thickness in structural magnetic resonance imaging data of 46 patients with SAD without comorbidities (except for depressed episode in one patient) compared with 46 matched healthy controls in a region of interest‐analysis and in whole‐brain. In a theory‐driven ROI‐analysis, cortical thickness was increased in SAD in left insula, right anterior cingulate and right temporal pole. Furthermore, the whole‐brain analysis revealed increased thickness in right dorsolateral prefrontal and right parietal cortex. This study detected no regions of decreased cortical thickness or brain volume in SAD. From the perspective of brain networks, these findings are in line with prior functional differences in salience networks and frontoparietal networks associated with executive‐controlling and attentional functions. Hum Brain Mapp 35:2966–2977, 2014. © 2013 Wiley Periodicals, Inc.

INTRODUCTION

Social anxiety disorder (SAD) is a common anxiety disorder [Kessler et al., 2010] with an estimated one‐year prevalence of 2.3% corresponding to more than 10 million affected persons in Europe [Wittchen et al., 2011, up to 6.5% in women and 4.8% in men, McLean et al., 2011] and a life‐time prevalence of about 10% [Kessler et al., 2010]. Genetic [Mosing et al., 2009; Stein and Stein, 2008] and neurobiological factors together play a role in the development of SAD. Neurofunctionally, anxiety‐related brain structures are more active and reactive in SAD including bilateral amygdaloid regions (extending into the (para)hippocampal area), bilateral insula, anterior cingulate cortex (ACC) and prefrontal cortical (PFC) structures [Etkin and Wager, 2007; Freitas‐Ferrari et al., 2010]. Patients with SAD exhibited this hyperactivity when confronted with negative faces and other social stimuli and also when processing non‐social emotional stimuli [Brühl et al., 2011; Shah et al., 2009]. Regarding structural‐anatomical brain changes, previous studies (Table 1) have provided mixed and in part contradictory findings. Mostly involved were amygdala, hippocampus, insula, orbitofrontal (OFC), PFC, temporal and occipital cortical structures [Frick et al., 2013; Irle et al., 2010; Liao et al., 2011; Syal et al., 2012; Talati et al., 2013]. When considering studies in other specific anxiety disorders, the findings are similarly non‐conclusive [Table 1, Hayano et al., 2009; Rauch et al., 2004; Uchida et al., 2008]. One reason for these results can be the rather small sample sizes (usually between 10 and 30 patients and likewise numbers of healthy subjects). One study with a total of 289 participants (156 patients with mixed anxiety disorders, including SAD, generalized anxiety disorder, panic disorder with notable comorbidities between the disorders, and 65 healthy control subjects [van Tol et al., 2010]) tried to overcome this problem. They primarily addressed common and differential changes in depression and anxiety disorders versus healthy participants and revealed reduced volumes in anxiety disorders in the left temporal cortex and the rostral ACC. Structural correlates of trait anxiety in healthy subjects were reflected in negative correlations of OFC [Kühn et al., 2011] and left amygdala volume [Blackmon et al., 2011] with trait anxiety, whereas trait anxiety was in another study positively correlated to amygdala and hippocampus volume [Baur et al., 2012].

Table 1.

Overview of structural differences in SAD (studies in related populations in grey)

Study	Methoda	Group	N b	Amy	HC	Ins	OFC	TC	PFC	Occ	Subc
Irle et al. [2010]	3T, man.	SAD	24/24	↓	↓	n.a.	n.a.	n.a.	n.a.	n.a.	ø
Liao et al. [2011]	3T, VBM	SAD	18/18	↓	↓	ø	ø	↓	↑	ø	ø
Syal et al. [2012]	3T, FS	SAD	13/13	ø	ø	↓	↓	ø	↓	↓	ø
Talati et al. [2013]	1.5T, VBM	SAD	33/37	ø	↑	ø	ø	↓	ø	ø	ø
Potts et al. [1994]	1.5T, man.	SAD	22/22	n.a.	n.a.	n.a.	n.a.	n.a.	n.a.	n.a.	øb
Frick et al. [2013]	1.5T, FACE	SAD	14/12	ø	ø	ø	ø	↑	ø	ø	ø
Hayano et al. [2009]	1.5T, VBM	PD	30/30	↓	↓	n.a.	n.a.	n.a.	n.a.	n.a.	n.a.
Uchida et al. [2008]	1.5T, VBM	PD	20/20	ø	ø	↑	ø	↑	↓c	ø	↑d
Rauch et al. [2004]	1.5T, FS	AP	10/20	ø	ø	↑	ø	↑	↑c	↑	ø
van Tol et al. [2010]	3T, VBM	Mixed	156/65	ø	ø	ø	ø	↓	↓c	ø	ø
Kühn et al. [2011]	3T, FS	Trait	−/34	ø	ø	ø	↓e	ø	ø	ø	↑e, f
Blackmon et al. [2011]	3T, FS	Trait	−/34	↓e	ø	ø	↑e	↑e	ø	ø	ø
Baur et al. [2012]	3T, FS	Trait	−/32	↑e	↑e	n.a.	n.a.	n.a.	n.a.	n.a.	øg

^aMagnetic field strength, analysis method; number of included subjects (patients/healthy subjects).

^bThalamus, putamen, caudatum.

^cACC.

^dClaustrum, midbrain, pons.

^eCorrelational analysis.

^fNcl. accumbens

^gCaudate.

↑, increased thickness/volume compared with controls; ↓, decreased thickness/volume compared with controls; ø, no significant difference; amy, amygdala (including parahippocampal gyrus); AP, animal phobia; FACE, fast accurate cortex extraction [Eskildsen and Ostergaard, 2006]; FS, Freesurfer; man, manual anatomical analysis; mixed, mixed anxiety disorders; n.a., not available (ROI‐based analysis); HC, hippocampus (including parahippocampal gyrus); Ins, insula; Occ, occipital cortex; OFC, orbitofrontal cortex; PD, panic disorder; PFC, prefrontal cortex (including anterior cingulate cortex); ROIs, regions of interest; SAD, Social Anxiety Disorder; Subc, subcortical structures; TC, temporal cortex (including temporal pole and other findings in the temporal lobe); Trait, trait anxiety (nonclinical, healthy participants); VBM, voxel based morphometry.

To overcome the small sample size problem of prior studies and the diagnostic heterogeneity of the study of van Tol et al., we examined cortical thickness in a rather large sample of patients suffering from SAD (without comorbid axis I‐disorders) compared with healthy participants. Our analysis focused on anatomically defined bilateral regions of interest (ROIs) derived from the prior studies, comprising amygdala, hippocampus and adjacent parahippocampal gyrus, anterior insula, ACC, OFC, fusiform gyrus (FFG), superior temporal gyrus and temporal pole. In addition, we computed a whole‐brain vertex‐wise analysis to search for differences besides these theory‐based predefined ROIs.

MATERIALS AND METHODS

Subjects

Participants were 50 outpatients with the current diagnosis of generalized SAD. In four patients, anatomical data were not analysable due to technical reasons (artefacts/defects in data). The remaining 46 patients were matched with 46 healthy control subjects (HCS) in terms of age and gender (see Table 2). All participants were right‐handed (Annett handedness questionnaire [Annett, 1970]). They had completed at least regular scholar education and showed no manifest cognitive impairments. Patients were recruited from the outpatient clinic at the Department of Psychiatry and Psychotherapy of the University Hospital Zurich before starting a group cognitive behavior therapy (CBT) for SAD. Prior psychological treatment (except for CBT) was not used as an exclusion criterion, because the ongoing symptoms and distress were severe enough for the patients to seek treatment again, pointing to a lack of success of the prior treatment. Healthy subjects were recruited via email‐lists and personal contact. In a clinical interview made by an experienced psychiatrist (A.B.B.) current and previous mental and neurological disorders were excluded. General anxiety level was measured in all subjects with the State‐Trait Anxiety Inventory (STAI, German version [Laux et al., 1981]). Social anxiety was assessed with the Social Phobia Scale and Social Interaction Anxiety Scale (SPS and SIAS, German version [Stangier et al., 1999]) and the Liebowitz Social Anxiety Scale (LSAS, German version) [Stangier and Heidenreich, 2005], concomitant depressiveness was assessed with Beck's Depression Inventory (BDI, German version) [Hautzinger et al., 1995]. Diagnoses of SAD and comorbid axis I‐diagnoses were made clinically by experienced psychiatrists and psychologists (MR, SW, and AD). Furthermore, the Mini‐International Neuropsychiatric Interview for DSM‐IV (M.I.N.I., German Version) [Ackenheil et al., 1999] was applied for confirmation. One patient fulfilled criteria for current depressive episode; SAD, however, was the primary diagnosis. All other patients had no axis I‐comorbidities and had no history of psychiatric and neurological disorders and head trauma. Exclusion criteria for all participants were pregnancy, excessive consummation of alcohol (>10 units/week), cigarettes (>2 packs/day), and caffeine (>10 cups/day), contraindications against magnetic resonance imaging (MRI). Control subjects were free of medication (except oral contraceptives) as determined in a semi‐structured clinical interview according to DSM‐IV.

Table 2.

Demographic and psychometric data

	HCS (N = 46)	SAD (N = 46)	Statistics
Gender	29 m, 17 f	29 m, 17 f	χ 2; P = 1.0
Age [Mean (SD, range)]	32.96 (8.87, 18–57)	33.13 (10.61, 19–53)	t = 0.084; P = 0.933
STAI [Mean (SD, range)]	31.3 (5.88, 21–55)	51.4 (10.8, 27–76)	t = −10.947; P < 0.001
LSAS [Mean (SD, range)]	n. a.	66.17 (20.4, 15–103)	n. a.
SPS [Mean (SD, range)]	n. a.	31.27 (16.12, 10–74)	n. a.
SIAS [Mean (SD, range)]	n. a.	37.66 (11.78, 9–60)	n. a.
BDI [Mean (SD, range)]	n. a.	16.12 (9.52, 0–41a)	n. a.

^aOne patient fulfilled criteria for current depressive episode, however, SAD was the primary diagnosis. All other patients were free of comorbid psychiatric and neurological disorders.

m, male; f, female; HCS, healthy control subjects; SAD, social anxiety disorder patients; STAI, state‐trait anxiety inventory—trait version; LSAS, Liebowitz Social Anxiety Scale; SPS, Social Phobia Scale; SIAS, Social Interaction Anxiety Scale; BDI, Beck's Depression Index.

Nineteen patients were taking antidepressant medication (selective serotonin reuptake inhibitors, selective serotonin/norepinephrine‐ reuptake inhibitors, mirtazapine, clomipramine, trazodone). The dosage was stable in all patients for at least four weeks prior examination according to the preinclusion interview. After complete description of the study, written informed consent was obtained. The study was approved by the local ethics committee and was conducted according to the Declaration of Helsinki.

Image Acquisition

MRI scans were acquired on a 3.0 T GE Signa HD Scanner (GE Medical Systems, Milwaukee) equipped with a transmit‐receive body coil and a commercial eight‐element sensitivity encoding (SENSE) head coil array. A volumetric three‐dimensional T1‐weighted fast spoiled gradient echo (FSPGR) scan was obtained with a measured spatial resolution of 0.94 × 0.94 × 1.00 mm³ (matrix 256 × 256 pixels, 172 slices, axial orientation) covering the whole brain. Further imaging parameters were: field of view (FOV) 240 × 240 mm², echo time (TE) 2.1 ms, repetition‐time (TR) 9.2 ms, inversion time (TI) 500 ms, flip‐angle 20°. Total acquisition time was about 6 min 20 s. Furthermore, T2‐weighted images were acquired to exclude possible T2‐sensitive brain abnormalities. For data quality check, all MRI scans were visually inspected. Four SAD datasets had to be excluded due to moderate to severe technical artefacts (defects, distortions).

Image Preprocessing

Cortical surface reconstruction, cortical parcellation, and subcortical volumetric segmentation were performed with the FreeSurfer image analysis suite (version 4.5.0), which is documented and available for download online (http://surfer.nmr.mgh.harvard.edu/). The technical details of these procedures are described in prior publications [Dale et al., 1999; Fischl and Dale, 2000; Fischl et al., 1999a,b, 2001, 2002, 2004a,b]. The three‐dimensional structural T1‐weighted MRI scans were used to construct models of each subject's cortical surface in order to measure cortical thickness. This fully automated procedure comprised segmentation of the cortical white matter [Dale et al., 1999], tessellation of the grey/white matter junction, inflation of the folded surface tessellation patterns [Fischl et al., 1999a,b] and automatic correction of topological defects in the resulting manifold [Fischl et al., 2001]. This surface was then used as starting point for a deformable surface algorithm designed to find the grey/white and pial (grey matter/cerebrospinal fluid (CSF)) surfaces with sub‐millimetre precision [Fischl and Dale, 2000]. The procedures for measuring cortical thickness have been validated against histological analysis [Rosas et al., 2002] and manual measurements [Kuperberg et al., 2003; Salat et al., 2004]. This method uses both intensity and continuity information from the surfaces in the deformation procedure in order to interpolate surface locations for regions in which the MRI image is ambiguous [Fischl and Dale, 2000]. For each subject, cortical thickness of the cortical ribbon was computed on a uniform grid (comprised by vertices) with 1 mm spacing across both cortical hemispheres, with the thickness being defined by the shortest distance between the grey/white and pial surface models. The thickness maps produced are not limited to the voxel resolution of the image and thus sensitive for sub‐millimeter differences between groups [Fischl and Dale, 2000]. The way in which the resolution of the cortical thickness maps goes beyond the resolution of the original acquisition is conceptually similar to a (conventional) partial volume correction procedure. The cortex is smooth at the spatial scale of a several millimetres, which is imposed as constraint by FreeSurfer to estimate the location of the surface with subvoxel accuracy. For instance, if a given voxel is darker than its neighbouring grey matter it probably contains more CSF and so the surface model is at a slightly different position than if the neighbouring voxels were brighter and therefore contain probably more white matter. Thickness measures were mapped to the inflated surface of each participant's brain reconstruction, this allowing visualization of data across the entire cortical surface (gyri and sulci) without the data being obscured by cortical folding. Data were resampled for all subjects and rendered onto a common spherical coordinate system [Fischl et al., 1999b]. Then a surface‐based vertex‐wise cortical thickness map was computed for each participant. For the whole‐brain vertex‐wise analysis, the data were smoothed on the surface tessellation using an iterative nearest‐neighbour averaging procedure with 615 iterations on the left hemisphere and 621 iterations on the right hemisphere, corresponding to a two‐dimensional surface‐based diffusion smoothing kernel with a FWHM of 30 mm.

In addition, the cerebral cortex was parcellated into units based on gyral/sulcal structure as implemented in FreeSurfer [Desikan et al., 2006; Destrieux et al., 2010; Fischl et al., 2004b] to generate ROIs for a priori hypothesis testing. We used the following parcellations as cortical ROIs (Table 3, Supporting Information Fig. S1): anterior insula, ACC, OFC, superior temporal gyrus and temporal pole, FFG, and additionally the parahippocampal gyrus. We separated superior temporal gyrus (STG) and temporal pole to increase spatial resolution, reduce noise, and due to their differential functional allocation (the temporal pole being associated with the (para)limbic system [Olson et al., 2007], the STG being involved in visuo‐spatial, auditive and language functions [Karnath, 2001]), and their different anatomical connections. The boundaries of these parcellations were described elsewhere [Destrieux et al., 2010]. Mean cortical thickness values within these ROIs were calculated from unsmoothed thickness maps.

Table 3.

Results of the ROI analysis

Anatomical region (indexa)	HCS, mean (SD)	SAD, mean (SD)	Statistics
			F(1,88)	P	d
(a) ROI analysis—cortical thickness (mm)
ACC L (6)	2.888 (0.247)	2.952 (0.243)	1.593	0.210	0.27
ACC R (6)	2.641 (0.238)	2.787 (0.265)	8.334	0.005	0.61
OFC L (24/64)	2.758 (0.164)	2.779 (0.196)	0.179	0.674	0.09
OFC R (24/64)	2.772 (0.162)	2.785 (0.195)	0.021	0.885	0.03
Insula L (18/47)	3.421 (0.210)	3.448 (0.228)	5.232	0.025	0.48
Insula R (18/47)	3.328 (0.222)	3.429 (0.234)	0.148	0.701	0.08
Temporal pole L (43)	3.268 (0.322)	3.343 (0.399)	0.836	0.363	0.19
Temporal pole R (43)	3.367 (0.3369	3.519 (0.342)	4.493	0.037	0.45
STG L (34)	3.150 (0.183)	3.172 (0.219)	0.158	0.692	0.08
STG R (34)	3.231 (0.162)	3.201 (0.235)	1.418	0.227	0.25
PHG L (23)	3.037 (0.198)	2.977 (0.227)	2.170	0.144	0.31
PHG R (23)	3.204 (0.264)	3.217 (0.267)	0.015	0.904	0.03
FFG L	3.019 (0.164)	3.023 (0.212)	<0.01	0.985	<0.01
FFG R	3.004 (0.184)	2.966 (0.233)	1.064	0.305	0.022
(b) ROI analysis—volume
Hippocampus L	4,868.1 (516.45)	4,739.6 (449.96)	2.571	0.112	0.33
Hippocampus R	4,767.6 (454.67)	4,760.3 (514.34)	0.036	0.850	0.04
Amygdala L	1,542.0 (195.20)	1,552.8 (186.56)	0.517	0.474	0.15
Amygdala R	1,556.9 (204.09)	1,558.1 (193.29)	0.160	0.690	0.08

Analysis including age as covariate of no interest.

^aIndex: anatomical region(s) according to Destrieux et al. [2010], see Supporting Information Figure S1. Statistics with medication as covariate: Supporting Information Table SI. Significant results when applying correction for multiple comparisons (P < 0.0098) in bold, results significant (P < 0.05) without correction for multiple comparisons in italics.

R right, L left, ACC anterior cingulate cortex, OFC orbitofrontal cortex, STG superior temporal gyrus, PHG parahippocampal gyrus, SAD social anxiety disorder, HCS healthy control subjects.

The subcortical segmentation procedure used to measure the volume of subcortical structures such as the amygdala and hippocampus takes into account three different kinds of information to disambiguate labels: (i) the prior probability that a given tissue class occurs at a specific location in the atlas, (ii) the likelihood of the image given that tissue class, and (iii) the probability of the local spatial configuration of labels given the tissue class. This latter term represents constraints on the space of allowable segmentations and prohibits label configurations that never occur in reality (e.g. the hippocampus is never located anterior to the amygdala). This technique has previously been shown to be comparable in accuracy to manual labeling [Fischl et al., 2002]. The segmentations were visually inspected for accuracy and none of the segmentations had to be excluded. Finally, the volumes of the subcortical structures as well as global brain measures were computed based on these segmentations.

Statistical Analyses of the ROIs

Differences in cortical thickness within the a priori selected regions (ROIs) were compared with an analysis of covariance (ANCOVA) with healthy controls versus SAD patients as between‐subject factor, and global mean cortical thickness and age as covariate of no interest. In the volumetric analyses, mean global volume was used instead of mean cortical thickness. Taking into account possible effects of antidepressant medication in the patients, we also computed an ANCOVA using additionally medication as covariate of no interest. To analyze correlations between the measures of social anxiety (LSAS, SIAS, SPS) we calculated bivariate correlations with the thickness/volume of the ROIs. ANCOVA, correlations, and psychometric statistics were done using SPSS 20.

Whole‐Brain Statistical Analyses

To detect local differences in cortical thickness between patients with SAD and healthy subjects, we computed vertex‐wise analyses using a general linear model with an initial height threshold of P < 0.05 fully corrected for multiple comparisons using 2,000 synthetic z‐score permutations (Monte Carlo simulations) on the cluster extent while simultaneously controlling for global mean cortical thickness.

In the resulting clusters, correlations between individual values and measures of social anxiety and an ANCOVA to detect effects of medication and age were calculated using SPSS 20.

RESULTS

Participants

We analyzed anatomical data from in total 46 patients with SAD (29 males, mean age 33 years, Table 2), which were matched on age and gender with 46 healthy participants (29 males, mean age 33 years). The groups differed significantly in trait anxiety (STAI‐T). The severity of social anxiety in the SAD group ranged between mild and severe, the mean scores of the SPS were slightly higher than those reported for SAD patients in the German validation study of this scale (mean = 29, SD 16). Degree of social anxiety (LSAS, SIAS, SPS) and general anxiousness (STAI‐T) were highly correlated (LSAS/STAI: r = 0.609, P < 0.0001, SIAS/STAI: r = 0.505, P = 0.001, SPS/STAI: r = 0.442, P = 0.006). The patients' degree of depression according to the BDI pointed to depressive symptoms, however without fulfilling the criteria for current major depressive episode (besides one patient). These symptoms were interpreted as secondary to the SAD symptoms and suffering.

ROI Analysis—Global Measures

There were no significant group differences in the global measures of total brain volume (t _(1,90) = 0.286, P = 0.776) and in the mean global cortical thickness (t _(1,90) = −0.347, P = 0.729). When adding medication as a covariate, there was no significant change (total brain volume: F _(1,89) = 0.053, P = 0.819, partial η ²(medication) = 0.010; mean global cortical thickness F _(1,89) = 1.094, P = 0.298, partial η ²(medication) = 0.023). However, due to the strong impact of the two global measures of brain volume we integrated them as covariates into the respective ANCOVAs (total brain volume for the volumetric analysis of the amygdala and hippocampus, mean global cortical thickness for the ROI‐based and whole‐brain cortical thickness analyses). Adding medication status as covariate brought no significant difference to the model without medication (partial η ² of the factor medication between 0.002 and 0.020, results of this analysis: Supporting Information Table SI). We found no relevant effect of gender on these results (partial η ² of the factor gender in those structures with significant group differences between 0.004 and 0.042, partial η ² of the interaction group × gender between 0.001 and 0.006).

ROI Analysis—Cortical Thickness

In the ROI analysis (Table 3, (a); Fig. 1), we found in SAD patients increased cortical thickness in the right temporal pole, the right ACC and the left anterior insula. The effect sizes of these differences were small to medium. Of these, only the anterior cingulate cortex cluster survived correction for multiple comparisons (P < 0.0098 uncorrected corresponding to P < 0.05 corrected) when using Bonferroni‐Sidak's correction taking into account the intercorrelation between the ROIs' cortical thicknesses (mean r = 0.34). There were no significant differences in the other ROIs (even at P < 0.10).

Figure 1.

Results of the ROI analysis. Given are mean and SEM. L left, R right, TP temporal pole, ACC anterior cingulate cortex, OFC orbitofrontal cortex, PHG parahippocampal gyrus, STG superior temporal gyrus. *Significant difference corrected for multiple comparisons (P < 0.0098), (*)significant difference without correction for multiple comparisons (P < 0.05).

Cortical thickness of the right anterior insula was positively correlated with LSAS (r = 0.357, P < 0.01) and with SPS (r = 0.342, P < 0.04), cortical thickness of the left parahippocampal gyrus was positively correlated with the SPS (r = 0.385, P < 0.02). There were no further significant correlations between measures of social anxiety and cortical thickness.

ROI Analysis—Subcortical Volumes

There were no significant group differences in amygdala and hippocampus [Table 3; (b)]. Correlations between measures of social anxiety and volume of amygdala and hippocampus were not significant.

Whole‐Brain Analysis—Cortical Thickness

The vertex‐wise whole‐brain analysis resulted in two clusters with significantly increased cortical thickness in SAD (Table 4, Fig. 2). One was located in the right middle frontal gyrus extending into superior frontal sulcus, belonging to the dorsolateral prefrontal cortex (DLPFC, Brodmann areas 6/8/9/46). The other covered right superior parietal lobule and angular gyrus and extended in part into right precuneus and inferior parietal lobule (Brodmann areas 7/39/19). We found no significant correlations between individual cortical thickness and LSAS, SPS, or SIAS.

Table 4.

Whole brain vertex‐wise group analysis of cortical thickness differences SAD versus HCS

	Size (mm2)	MNI coordinates x/y/z	Cluster wise P (corrected)
(a) Contrast SAD > HCS
Anatomical region (BA)
Superior/inferior parietal lobe/precuneus R (7/19)	3,366	36/−64/45	0.0105
Middle frontal gyrus R (6/8/9)	2,700	37/28/29	0.0405
(b) Contrast HCS > SAD none

Brain regions with significant differences of cortical thickness between SAD and healthy control subjects (HCS), vertex‐wise P < 0.05, corrected for multiple comparisons using a Monte Carlo simulation, resulting in a corrected cluster‐wise P < 0.05.

Figure 2.

Results of the vertex‐wise whole‐brain cortical thickness analysis (significant group differences, P < 0.05 corrected for multiple comparisons).

Furthermore, we calculated a comparison between patients currently taking medication and patients free of medication (data not shown) which revealed no significant differences in the whole brain analysis (P < 0.10).

DISCUSSION

This study identified increased cortical thickness in SAD in right DLPFC and right parietal cortex in the whole‐brain analysis and in right ACC, right temporal pole and left anterior insula in the ROI analysis. There were no brain regions with decreased cortical thickness in SAD, nor volumetric differences in amygdala and hippocampus between SAD patients and healthy controls.

When comparing our results with previous studies on anatomical differences in SAD, there is one clear difference: We find solely brain regions with increased thickness and cannot confirm previous findings of decreased cortical thickness or brain volume. This is in line with the two most recent studies [Frick et al., 2013, Talati et al., 2013], but not with prior studies which detected mostly reduced brain volumes and cortical thickness measures (Table 1). Our study included a markedly larger sample than the others which should have given our study the power to detect at least moderate differences in both directions. However, one reason for this differing finding could be a methodological one, as some of the prior studies used manual tracing of amygdala and hippocampus or voxel‐based morphometric methods, whereas our study used the FreeSurfer software suite, a well‐validated and automated surface‐based cortical reconstruction and parcellation method as well as subcortical segmentation procedure. The method used in our study has been shown to be reliable compared with manual and histological parcellation and segmentation methods [Kuperberg et al., 2003; Morey et al., 2009; Rosas et al., 2002; Salat et al., 2004]. In addition, cortical thickness‐based methods might be more sensitive in detecting structural difference than volume‐based methods [Hutton et al., 2009]. However, Syal et al. used Freesurfer as well and found (in a sample of 13 SAD patients vs. 13 healthy participants) only decreased cortical thicknesses [Syal et al., 2012], such that the analytical method is presumably not the only reason for this discrepancy between our data and parts of the literature. On the other hand, FreeSurfer could be less sensitive in subcortical structures than manual or multi‐atlas based tracing methods [Hanson et al., 2012]. Therefore, the results in cortical regions in our study might have more power than those in subcortical regions. Further methodological studies on discrepancies between different methods are strongly needed.

Other reasons for this difference could lie in the examined subjects: Our patient sample was slightly more strictly selected than others, the patients had no comorbid and other lifetime diagnoses (except for one patient fulfilling criteria of current depressive episode, which we interpreted as reactive due to suffering of SAD symptoms). Furthermore, our sample has a mild male preponderance. This, however, is rather typical for SAD, where the gender‐ratio more balanced than female prone, as it is in other anxiety disorders [McLean et al., 2011]. Indeed, other studies had such a male preponderance as well [Liao et al., 2011; Potts et al., 1994; Syal et al., 2012; however, Talati et al., 2013]. Regarding symptom severity, age and education level, participants in our study were not different from the other studies. However, compared with other studies, our study included also patients currently taking antidepressants which were excluded in the study of Irle et al., but not in others. Antidepressants have been shown to normalize decreased hippocampal volume in depression [Frodl et al., 2008; however, no effect: Vythilingam et al., 2004]. However, adding medication as factor into the ANCOVA made no difference to the results; we found particularly no decreased volume of amygdala and hippocampus. Thus, in summary, we can at the current state of research in structural analysis methods not clearly explain the differences between our results and those of others. However, a strength of our study is the rather large sample size together with a representative selection of patients, which yields our results quite reliable and significant.

Another methodological aspect is the correction for multiple comparisons in the hypothesis‐ and theory‐driven ROI analysis. Compared with other statistical situations ROIs in the brain are due to biological reasons not independent, but at least moderately intercorrelated, which makes correction for multiple comparisons difficult and poses the risk of increasing Type II‐error markedly [Lieberman and Cunningham, 2009]. Therefore, and due to the a priori positioning of the ROIs we consider those findings which are significant at an uncorrected level, but do not survive correction for multiple comparisons, as relevant and discuss them along with the corrected results.

From a general functional perspective, those brain region with increased thickness in SAD fit into models of hyperactive neural circuits involved in the processing of emotional stimuli (temporal pole, anterior insula) and of frontoparietal networks particularly associated with executive, emotion regulating and attentional functions (DLPFC, ACC, parietal lobe) [e.g. Etkin et al., 2009; Ochsner et al., 2012]. Particularly the increased thickness in the DLPFC and the superior parietal cortex in SAD are in line with models of disturbed, overactive and dysregulated attentional networks in anxiety disorders [Sylvester et al., 2012], which are predominantly right lateralized, similar to our findings.

The lack of correlation with symptom severity in our study could further suggest that these two regions are rather a correlate of the category of disorder than of any psychopathological dimension. On the other hand, particularly the higher thickness of the DLPFC as well as the ACC could both be the result of continuous efforts of coping or attempts of emotion regulation, thus rather a compensatory increase of thickness. Both, the DLPFC and the ACC have been shown to be implicated in emotion regulation circuits in many studies [e.g. Herwig et al., 2007; Ochsner et al., 2002; meta‐analyses: Diekhof et al., 2011; Kalisch, 2009]. The right DLPFC is furthermore known to be more active during emotion suppression [Goldin et al., 2008; Phan et al., 2005], which is used more frequently in SAD patients [Kashdan and Steger, 2006], which would support the interpretation of compensatory increase of DLPFC thickness in the current study.

Thicknesses of anterior insula and temporal pole were positively correlated with the severity of SAD symptoms. This supports is in line with typical functional models of SAD and other anxiety disorders which implicate an increased activity and reactivity of amygdala and anterior insula in response to threatening or feared stimuli [meta‐analysis: Etkin and Wager, 2007], but also to non‐specific general negative emotional stimuli [Brühl et al., 2011; Shah et al., 2009]. Increased responsiveness of the insula is interpreted as link between internal, particularly bodily, and external information [Craig, 2009; Paulus and Stein, 2010] which is related to anxiety and anxiety disorders. Reduced cortical thickness in the temporal pole has been shown bilaterally in a group of violent offenders ranking high in psychopathy, which is considered as the opposite of anxiety and particularly of social anxiety [Gregory et al., 2012]. Together with the findings in our study, this finding supports a role of the temporal pole region for anxiety, perhaps with a specific focus on social aspects of anxiety.

In addition to this increased propagation of emotional information, models have suggested, that a diminished regulation of these bottom‐up processing regions by prefrontal brain regions might play a role. Some studies found an increased DLPFC activity [Klumpp et al., 2012], but disturbed connectivity [Hahn et al., 2011; Liao et al., 2011; Prater et al., 2013] and reduced structural connections [Baur et al., 2013, 2011; Phan et al., 2009] are frequently reported, which could be interpreted as either a functional disturbance of the DLPFC or rather disturbed regulatory connections between these regulating regions and the subcortical and limbic targets of regulation. Our data would, at least from the cortical perspective, support such models of disturbed regulatory connectivity which is reflected by increased thickness in both, the hyperactive limbic and the compensatorily, but unsuccessfully hyperactive prefrontal structures.

In addition, prior studies have identified increased activations in brain regions within the dorsal attention network [e.g. Brühl et al., 2011].

In summary, comparing the pattern of structural differences of the current study with functional models of SAD, we found overlapping networks, such salience processing regions (insula), executive control regions (DLPFC, cingulate) extending into the dorsal attention network (DLPFC, superior parietal cortex). However, particularly subcortical regions show consistently altered functions, but no structural differences in our study. This discrepancy could reflect disturbed connections between subcortical circuits and regulating cortical structures as has been shown on the functional and structural level [Baur et al., 2013, 2011; Hahn et al., 2011; Liao et al., 2011; Phan et al., 2009].

However, when trying to transfer functional findings and models to the here presented structural findings, one has to take into account that the relation between cortical thickness and function is not totally clear: When a brain region is clearly atrophic, with a marked reduction of cortical thickness, this is strongly related to diminished function of a region, for instance in degenerative disorders [Julkunen et al., 2010; Kallianpur et al., 2012] or mental retardation [Zhang et al., 2011]. In parallel, practicing a specific skill is associated with increased thickness of respective regions in rats [Anderson et al., 2002] and humans [Engvig et al., 2011; Wenger et al., 2012], particularly in younger participants. In parallel, professional golf players had higher grey matter volumes compared with less‐skilled golfers [Jäncke et al., 2009], whereas in professional ballet dancers, length of training or specific skill ability were negatively correlated with grey matter volume in training‐ and skill‐related brain structures [Hänggi et al., 2010]. In other domains again, performance and cortical thickness were correlated also in other domains [Liem et al., 2012; Yin et al., 2011]. In summary, structural and functional or behavioural differences cannot be directly transferred into each other conclusively, particular not causally. However, in some brain regions such as in insula, temporal and prefrontal cortex, the data from functional and structural studies (increased thickness and increased activity) in SAD are rather consistent and point into a common direction.

The question remains, wether the described differences in SAD compared with HCS are anatomical reasons of the disorder or reflect potential compensatory effects. One could argue that the changes in anterior insula and temporal pole are caused by or correspond to anxious hyper‐arousal and hyper‐reactivity, paralleled by the correlation with symptoms, which then coincide in the ACC with increased, possibly compensatory regulatory efforts, involving prefrontal regions such as MPFC and DLPFC. The increased thickness in the parietal cortex corresponds to increased activations in functional studies [Brühl et al., 2011] and early information processing biases in SAD [Miskovic and Schmidt, 2012a] and could reflect hyperfunctional, but dysregulated attention networks [Sylvester et al., 2012] and disturbed connections in regulatory circuits [Baur et al., 2013]. These findings extend the current model of SAD [Freitas‐Ferrari et al., 2010; Miskovic and Schmidt, 2012b] by emphasizing attention and visual‐perception related brain regions.

Limitations and Methodological Issues

A limitation of the current study can be seen in the lack of measures of intelligence or IQ and respective matching. Total brain volume or grey matter volume have been shown to correlate with intelligence [review: Gray and Thompson, 2004], with moderate correlations between 0.21 and 0.39 [McDaniel, 2005], most prominent in frontal brain regions [e.g. Colom et al., 2013]. Some volumetric studies therefore measure and match IQ. However, this was not done in our study. We used therefore scholar education as minimal proxy for intelligence and to exclude relevant cognitive impairments. Nearly all subjects were currently employed, in training or going to college or university, which is comparable to the other mentioned studies. Furthermore, there are, to the best of our knowledge, no clear data in the literature that patients suffering from SAD have specific intellectual impairments [Pubmed search, Bourke et al., 2012], which exceed possible testing‐stress related disturbances [O'Toole and Pedersen 2011].

As another limitation could be mentioned, that we did not assess life‐time and duration of current medication in the patient group. The requirement was that the medication was stable for at least four weeks prior inclusion into the study. However, we tested in the ANCOVA and in a group comparison between patients with and without medication for possible effects of pharmacological treatment. We detected no relevant influence of medication in our group.

Another potential limiting aspect is a methodological one, typical for most, if not all brain imaging studies: Researchers have to balance their analyses between on the one hand statistical thresholds resulting in multiple results which are then potentially biased by the problem of false positive findings due to multiple comparisons. On the other hand, reducing multiple comparisons by using predefined ROIs (as was done in this study) has the disadvantage, that cortical thickness is averaged within the ROIs which could obscure possibly existing differences and bias results towards the null hypothesis [Poldrack and Mumford, 2009], similar to the use of too strict statistical thresholds when correcting for multiple comparisons with any one method which then increase Type II errors again [for a detailed discussion of these problems refer to for instance, Button et al., 2013; Lieberman and Cunningham, 2009]. Therefore, our results cannot exclude effects in some of the examined ROIs. However, we have tried to balance as far as possible and reasonable between these two aspects (reporting false positive results by using too low thresholds versus loosing relevant findings by applying too strict thresholds) by combining an a priori, theory‐driven ROI approach, where we report those results surviving a correction for multiple comparisons as well as those fulfilling merely the uncorrected statistical threshold, with a more exploratory whole‐brain approach where we apply a rather rigorous correction for multiple comparisons.

Future Perspective

Future studies including and overarching multiple disorders, for instance from the whole group of anxiety disorders could help clarifying (a) common and specific factors for the respective disorders and (b) also effects of comorbidity compared with patients with only one disorder. Furthermore, the role of sensory and/or sensory‐integrating structures in SAD is until now more or less underinvestigated and under‐discussed. Clarifying sensory‐integrating functions associated with the dorsal attention network, perhaps also deficits or dysregulations, in SAD could provide support for therapeutic approaches aiming at attention modification [Hakamata et al., 2010; Heeren et al., 2011, 2012; Neubauer et al., 2013].

Furthermore, methodological research on the influence of different parameters of data acquisition and analysis, including software, could help developing more generally comparable methods and reducing the number of (seemingly) contradictory studies.

CONCLUSION

In conclusion, our study revealed increased thickness in prefrontal, parietal and insular cortex in SAD, but no decreases of cortical thickness or brain volume. These results fit with recent findings on the functional level, pointing more toward distributed differences in SAD on the level of networks such as the prefrontal regulatory and the dorsal attention network than to circumscribed disturbances focusing, for instance, on the amygdala.

Supporting information

Supporting Information