Article Text

Download PDFPDF

Research paper
A smaller amygdala is associated with anxiety in Parkinson’s disease: a combined FreeSurfer—VBM study
  1. Chris Vriend1,2,3,4,
  2. Premika SW Boedhoe1,2,3,
  3. Sonja Rutten1,2,
  4. Henk W Berendse3,4,
  5. Ysbrand D van der Werf2,3,5,
  6. Odile A van den Heuvel1,2,3
  1. 1Department of Psychiatry, VU University Medical Center, Amsterdam, The Netherlands
  2. 2Department of Anatomy & Neurosciences, VU University Medical Center, Amsterdam, The Netherlands
  3. 3Neuroscience Campus Amsterdam, VU/VUMC, Amsterdam, The Netherlands
  4. 4Department of Neurology, VU University Medical Center, Amsterdam, The Netherlands
  5. 5Netherlands Institute for Neuroscience, Royal Netherlands Academy of Arts and Sciences, Amsterdam, The Netherlands
  1. Correspondence to Chris Vriend, Department of Psychiatry, VU University Medical Center, Medical Faculty Building, Department of Anatomy and Neuroscience, room G102-b, van der Boechorststraat 7, Amsterdam 1081 BT, The Netherlands; c.vriend{at}


Background Up to 50% of all patients with Parkinson's disease (PD) suffer from anxiety symptoms, a much higher percentage than in the general population. This suggests that PD associated pathological alterations partly underlie these symptoms, although empirical evidence is limited.

Methods Here we investigated the association between anxiety symptoms measured with the Beck Anxiety Inventory (BAI) and hippocampal and amygdalar volume in 110 early-stage patients with PD. Measures of anxiety in PD are often obscured by overlap with the somatic symptoms. We therefore also used a subscale of the BAI, established by our recent factor analysis, that reflects ‘psychological’ anxiety symptoms and is independent of the severity of PD-related motor and autonomic symptoms. We used FreeSurfer and voxel-based morphometry for the volumetric analyses.

Results Both software packages showed a negative correlation between the ‘psychological’ subscale of the BAI, but not total BAI and volume of the left amygdala, independent of the severity of motor symptoms, autonomic dysfunction and dopaminergic or anxiolytic medication status.

Conclusions These results confirm studies in non-PD samples showing lower left amygdalar volume in anxious patients. The results also indicate that the ‘psychological’ BAI subscale is a better reflection of neural correlates of anxiety in PD. Whether the left amygdalar volume decrease constitutes a premorbid trait, a PD-associated neurobiological susceptibility to anxiety or arises as a consequence of chronic anxiety symptoms remains to be determined by future prospective longitudinal studies. Nonetheless, we speculate that the Parkinson pathology is responsible for the reduction in amygdalar volume and the concomitant development of anxiety symptoms.

  • MRI
View Full Text

Statistics from


Anxiety symptoms are very common in Parkinson’s disease (PD), affecting as many as 50% of all patients1 and yet surprisingly little research has been devoted to its pathophysiology. Anxiety in PD is associated with poorer appreciation of the quality of life and has, among others, been linked to more severe motor symptoms, autonomic dysfunctions and severity of depressive symptoms; see ref. 2 for a review.

The severity of anxiety symptoms in PD has previously been associated with reduced striatal dopamine transporter (DaT) availability,3 although increased availability has also been reported.4 Lesioning of the substantia nigra pars compacta in rodents leads to anxiety-like behaviour that is reversed by dopamine agonist infusion,5 suggesting a causal role for dopamine dysfunction in the development of anxiety symptoms. Nevertheless, studies on whether dopamine replacement therapy can alleviate anxiety symptoms in patients with PD have been inconsistent6 ,7 and dysfunction of other neurotransmitters, such as serotonin and noradrenalin, are probably also involved in the development of anxiety in PD.

The amygdala is critical for fear conditioning,8 while the hippocampus serves an important role in the declarative memory and the encoding of contextual representations.9 Studies in non-PD samples have shown an important role for the hippocampus and amygdala in the pathophysiology of anxiety disorders.10 Functional imaging studies have consistently shown hyperactivation of the amygdala during symptom provocation in patients with an anxiety disorder.10 Hippocampal activity is also altered, especially in patients with post-traumatic stress disorders (PTSD), but there are inconsistencies in the direction of the effect.11 ,12 A meta-analysis of structural studies in PTSD showed that patients with PTSD have smaller hippocampal and amygdalar volume than healthy controls.13 Reductions in hippocampal and amygdalar volume have also generally been observed in patients with panic disorder,14 social anxiety15 and generalised anxiety disorder,16 although volume increases have also been reported.17 To the best of our knowledge, no morphological studies have so far been performed on anxiety symptoms in PD. In a previous study by our own group, however, we observed a negative correlation between depressive symptoms and bilateral hippocampal and right amygdalar volume.18 Depression and anxiety are frequently co-occuring in PD and may have partly overlapping underlying pathophysiological mechanisms.19 Atrophy of the bilateral amygdala was also observed in patients with PD with mild depression compared with healthy controls.20 Other studies found mainly cortical differences between patients with PD who were depressed and non-depressed, see21 for a review.

In this study we investigated the volumetric correlates of anxiety symptoms in a large sample of patients with PD. Based on previous findings in patients without PD with anxiety and in patients with PD with depression we hypothesised that anxiety symptoms would be associated with reduced volume of the amygdala and hippocampus. Measuring the severity of anxiety symptoms in PD is obscured by the overlap with PD-related somatic symptoms (eg, motor and autonomic symptoms).2 The clinimetric properties of the existing anxiety rating scales are therefore less suitable to diagnose anxiety in PD.1 To disentangle anxiety symptoms from PD-related motor impairments and autonomic dysfunctions we previously conducted a factor analysis on the Beck Anxiety Inventory (BAI) and showed that the BAI can be partitioned into an affective subscale, reflecting ‘psychological’ anxiety symptoms, that is unrelated to motor and autonomic failure and four somatic subscales.22 In the current study, we also correlated volume with these subscales and hypothesised that amygdalar and hippocampal volume would show the strongest association with psychological symptoms of anxiety.

Materials and methods


Patients were consecutive cases diagnosed by a movement disorder specialist according to the UK PD Brain Bank criteria for idiopathic PD between May 2008 and October 2012 at the outpatient clinic for movement disorders of the VU University Medical Center (VUMC; Amsterdam, the Netherlands). Data from these patients were pooled with data from an additional seven medication-naïve patients with PD of a previous fMRI study23 of whom there was not already data available in the database from the outpatient clinic. All participants provided written informed consent according to the declaration of Helsinki and the study was approved by the local research ethics committee.

Clinical measures

In all patients, we used the Unified Parkinson's Disease Rating Scale motor section (UPDRS-III) and Hoehn and Yahr stage to assess disease severity and disease stage, respectively. Severity of anxiety symptoms were assessed with the BAI and depressive symptoms with the Beck Depression Inventory (BDI). We measured severity of autonomic dysfunction with the Scales for Outcomes in Parkinson's disease—Autonomic (SCOPA-AUT). Patients were excluded if they missed more than 1/6th (16.7%) of the items on the BAI. Otherwise we used mean imputation to fill in the missing items. If possible, mean imputation was also applied to the BDI and SCOPA-AUT items. On these scales, however, >16.7% missing items did not lead to exclusion of the patient, only to a missing value on the total score of that particular scale. No imputation was used for the UPDRS-III. Further exclusion criteria were signs of dementia (Mini Mental State Examination (MMSE) ≤24) and gross brain pathology or movement artefacts visible on the MRI. Patients were also excluded if their imaging and clinical data were not acquired on the same date. Dopamine replacement therapy dosages were converted to levodopa equivalent daily dose (LEDD) as described previously.24

BAI subscales

The previously conducted principal component analysis of the BAI resulted in five subscales, corresponding to one affective and four somatic symptom dimensions (thermoregulation, tremble, hyperventilation and hypotension).22 The affective subscale was unrelated to the severity of motor or autonomic symptoms and we therefore considered this to be a more reliable measure of the psychological anxiety symptoms in PD than the total BAI score. Online supplementary table S1 lists the BAI items that belong to each BAI subscale.

Image acquisition

All imaging data were collected on a GE Signa HDxT 3 T MRI scanner (General Electric, Milwaukee, USA) at the VU University Medical Center (Amsterdam, the Netherlands). We acquired structural images using a three-dimensional (3D) T1-weighted sequence with an eight-channel head coil (matrix 256×256, field of view=25 cm, TI=450 ms, TE=3.004 ms, voxel size 1×0.937×0.937 mm, 172 slices).

FreeSurfer preprocessing

We used FreeSurfer 5.3 to automatically segment the left and right amygdala and hippocampus and calculate their volume. Automated segmentation of the amygdala and hippocampus using FreeSurfer has an accuracy comparable to manual tracing.25 The brain parcellation and segmentation were run using the standard ‘recon-all’ script and all settings were left at default. The output was thoroughly inspected for segmentation errors. Quality checking and volume calculation was aided by scripts supplied by ENIGMA (Enhancing Neuro-Imaging Genetics Through Meta-Analysis; Intracranial volume (ICV) was also calculated to correct for interindividual differences in total brain size. All volume measures were exported to IBM SPSS V.20 (Armonk, New York, USA) for statistical analyses.

Voxel-based morphometry preprocessing

To corroborate the results from the FreeSurfer analyses we also conducted voxel-based morphometry (VBM) analyses in SPM8 (Wellcome Trust Center for Neuroimaging, London, UK) using the VBM8 toolbox ( Images were reoriented to the anterior/posterior commissure axis and segmented into grey-matter, white-matter and cerebrospinal fluid with a bias-field correction cut-off of 30 mm full width at half-maximum (FWHM). Next a group-specific template was created using high-dimensional DARTEL (Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra; default settings) and flow fields were calculated that contain information on the non-linear deformations between each subject’s scan and the DARTEL template. The resulting flow fields were subsequently applied to each subject’s grey matter images to warp them to Montreal Neurological Institute (MNI) space using linear affine transformation and non-linear deformation with the parameters obtained with DARTEL. Grey matter images were modulated to preserve relative regional volumes and correct for individual differences in brain size. Finally, the segmented, modulated and normalised images were smoothed using a 10 mm FWHM Gaussian kernel. Quality checks on the segmented and normalised images, such as segmentation errors and sample homogeneity, were performed with the VBM8 tools.

Data analyses

Correlations between clinical measures, demographics and volume were analysed with Pearson's r or Spearman's r (rs) correlation coefficient, depending on the distribution of the variable.

Hierarchical multiple regression analyses were performed to examine the relation between BAI (sub) score and amygdalar or hippocampal volume determined with FreeSurfer. Age, gender and ICV were simultaneously added as nuisance regressors to the first block of all regression models. BAI total score was added to the second block of the model to examine the association between anxiety symptoms and volume of the four regions-of-interest (ROIs), above and beyond the effects of age, gender and ICV. A total of four models with the BAI total score was constructed with left and right amygdalar or hippocampal volume specified as the dependent variables (figure 1).

Figure 1

Schematic overview of the backward stepwise regression selection procedure. Coefficients for the final models were selected by eliminating them one by one from the model according to the significance of their β (p>0.05). This procedure continued until all coefficients in the model had a β with p<0.05. Age, gender and ICV were always retained in the model irrespective of their significance. The selection procedure was carried out for volume of the left and right amygdala and hippocampus. See the method section for more information. (Inlay) Regression models with the Beck Anxiety Inventory (BAI) total as coefficient of interest and age, gender and ICV as nuisance covariates. Regression analyses were carried out for volume of the left and right amygdala and hippocampus. Abbreviations: DV, dependent variable; ICV, intracranial volume.

For multiple regression analyses with the BAI subscales we added an interim step to limit the number of tests that had to be performed (ie, multiple testing problem). For all four ROIs, age, gender and ICV were simultaneously added as nuisance regressors to the first block. To determine which scales had to be modelled with volume of the four ROIs, we added the score on the BDI and all five BAI subscales to the second block for a backward selection procedure. See figure 1 for a schematic representation of this procedure. BDI and BAI subscales were removed step-by-step from the model if their β had a p>0.05. The model was subsequently rerun without this (sub) scale and p values were re-evaluated. This process was repeated until the β's of all remaining scales had a p<0.05. Owing to their known influence on regional volume, age, gender and ICV were always retained in the model irrespective of their significance. Four final models (one for each ROI) were selected by the procedure and statistically evaluated. The left amygdala was modelled as dependent variable with ‘affective’ subscale as independent variable, the right amygdala (dependent variable) with the ‘hyperventilation’ subscale (independent variable) and the left and right hippocampus (dependent variables) with the ‘hypotension’ subscale (independent variable; see figure 1). Importantly, the backward selection procedure removed BDI score from all models. We ensured that all models met the assumptions of multiple regression analyses, including normality of the residuals, multicollinearity and homoscedasticity. Regressors were considered significant if they fell below an α of p<0.018. This critical value was established with SISA (, which uses the mean correlation between variables (ROI volumes) that are mutually correlated (mean r=0.51, eight comparisons) for the α correction and allows one to perform a less stringent correction than the classical Bonferroni method for multiple comparisons.

In VBM, statistical analyses were performed in the context of the general linear model. Regression models that showed a significant association between the BAI subscale and FreeSurfer volumes were reproduced in VBM. These ROIs were based on the Automated Anatomical Labelling (AAL) atlas and created with the Wake-Forest University PickAtlas tool 3.0. BAI total score was modelled with grey matter volume of all four ROIs. Regression coefficients were determined by drawing 3.5 mm spherical ROIs around the peak voxel of significant clusters using MarsBaR, (, extracting its mean grey-matter estimate to run multiple regression analysis with their respective BAI (sub)score in SPSS. The ICV, age and gender were added as nuisance regressors to all VBM analyses. Results were considered significant if they fell below an α of p<0.05, family-wise error (FWE) corrected for multiple comparisons.


Demographics and clinical data

Data from 111 patients with PD remained available for analyses. See the flow chart in figure 2 for an overview of the patients with PD that met our exclusion criteria. One additional patient was excluded due to a segmentation failure in FreeSurfer. Table 1 lists the demographic and clinical data. Thirteen patients were using anxiolytics or mood-stabilizers. The mean BAI total score of the total sample was 12.3±8.3. BAI total (r=0.25, p=0.009), but not the affective subscale (rs=0.12, p=0.22), correlated positively with UPDRS-III. The affective subscale also showed a lower association with autonomic symptoms (rs=0.19, p=0.05) than the BAI total score (BAI total: r=0.52, p<0.001). BAI total score or subscale scores did not correlate with age or LEDD.

Table 1

Sample characteristics

Figure 2

Flow of patients included in this study. FS, FreeSurfer; VBM, voxel-based morphometry.

FreeSurfer analyses

The multiple regression analyses showed that there was no relation between BAI total score and volume in all four ROIs beyond the effects age, gender and ICV. The ‘affective’ subscale showed a negative association with volume of the left amygdala (β=−0.24, p=0.001; see figure 3A) and the ‘hypotension’ subscale with volume of the right hippocampus (β=−0.19, p=0.012; see table 2). Although the β's of the ‘affective’ and ‘hypotension’ subscales in these models were modest, their addition to the model with only age, gender and ICV significantly increased the amount of explained variance by the model (‘affective’: ΔR2=0.05, p=0.001; ‘hypotension’: ΔR2=0.04, p=0.01). The ‘hypotension’ subscale was also associated with left hippocampal volume (β=−0.15, p=0.05) and the ‘hyperventilation’ subscale with right amygdalar volume (β=−0.16, p=0.025), but these associations fell short of our predefined statistical threshold (p<0.018).

Table 2

Multiple regression analyses on FreeSurfer volumes

Figure 3

Regression of the ‘affective’ subscale with volume of the left amygdala determined by FreeSurfer (A) and voxel-based morphometry (VBM) (B). (A) Partial plot of the association between volume of the left amygdala—as determined with FreeSurfer—and the ‘affective’ subscale while correcting for age, gender and intracranial volume (ICV). (B) Cluster of grey-matter volume in the left amygdala—as determined with VBM—correlating negatively with the ‘affective’ subscale. The VBM analysis was also corrected for age, gender and ICV.

As a post hoc analysis we also added UPDRS-III, SCOPA-AUT, LEDD or lateralisation of motor symptoms to the models to exclude the possible influence of additional nuisance regressors on the association. Adding UPDRS-III, SCOPA-AUT, LEDD or lateralisation of motor symptoms to the model had no effect on the association between the ‘affective’ subscale and left amygdalar volume (all p’s<0.018). In contrast, after adding UPDRS-III score to the association between the ‘hypotension’ subscale and left hippocampal volume this association was no longer significant (β=−0.18, p=0.03). We also performed the analyses in a subgroup of patients that did not use anxiolytics (benzodiazepines, tricyclic antidepressants, serotonin reuptake inhibitors (SSRIs)) or mood-stabilizers (lithium). Exclusion of these patients did not affect the association between the ‘affective’ subscale and left amygdalar volume (β=−0.21, p=0.007), but did affect the association between the ‘hypotension’ subscale and right hippocampal volume (β=−0.17, p=0.04).

VBM analyses

To confirm the above reported results we conducted the same analyses with VBM. Similar to the FreeSurfer analyses, there was no association between BAI total score and grey matter volume in the left and right hippocampus or the right amygdala. There was, however, a negative association between BAI total score and volume of the left amygdala (ke=149, PFWE=0.018, β=−0.27, p=0.003; see table 3 and see online supplementary table S2). In agreement with the FreeSurfer analysis, there was also an association between the ‘affective’ subscale and left amygdalar volume (ke=168, PFWE=0.012, β=−0.27, p=0.003; see figure 3B). No association was observed between the ‘hypotension’ subscale and right hippocampal volume.

Table 3

Correlation between beck anxiety inventory (BAI; sub) scales and voxel-based morphometry volume


In this study, we showed that symptoms of anxiety in patients with PD, and ‘psychological’ symptoms of anxiety in particular, show a negative correlation with volume of the left amygdala. This association was not affected by the severity of motor or autonomic symptoms or medication status. Correlations with hippocampal volume did not survive statistical thresholding. Our results are consistent with findings in anxious patients without PD.13–15 The amygdala is critically involved in emotional processing and fear conditioning. Neuroimaging studies in healthy individuals have shown that pharmacologically induced fear, fearful stimuli and emotional faces all elicit amygdalar activation; see ref. 26 for a review. These results are mirrored by findings on fear learning in animal models.8 Anxiety disorders, such as panic disorder, generalised anxiety disorder, PTSD, social and specific phobia, all seem to be characterised by an increased reactivity of the amygdala10 and normalisation of this activity correlates with symptom improvement after treatment.27 No functional neuroimaging study has yet been performed on anxiety in patients with PD. Patients with PD without anxiety do, however, show a blunted amygdala activation in response to fearful stimuli compared with healthy controls (see ref. 28 for a review). How these deficits in emotional processing relate to the development of anxiety symptoms remains to be determined.

The association between psychological anxiety symptoms and amygdalar volume was lateralised to the left. This result was not due to lateralisation of motor symptoms—and the presumed increased pathological load in the contralateral hemisphere—since adding this variable to the multiple regression model had no effect on the reported result. No differences in amygdalar volume or anxiety symptoms between left, right or bilaterally affected patients occurred (data not shown). Multiple studies have shown that the processing of negative stimuli by the amygdala and amygdalar dysfunction in anxiety disorders is left-lateralised (see refs. 13, 29 and 30 for reviews). Reports on this hemispheric asymmetry may in part be due to differences in methodology, valence of the stimuli (in the case of fMRI studies) or anatomical asymmetry.31 Nevertheless, it has been proposed that the left amygdala is more involved in detailed emotional information processing and detecting stimulus arousal, while the right amygdala is more important for rapid and automatic stimulus detection.32 No information is yet available on what the neurobiological mechanism is behind decreases in left amygdalar volume, dysfunction of detailed emotional information processing and the development of anxiety. This is partly because most studies on anxiety, including the present, are cross-sectional and we can therefore not draw any firm conclusions on whether decreased amygdalar volume constitutes a neurobiological susceptibility to anxiety or that amygdalar shrinkage is the consequence of chronic anxiety symptoms. The amygdala undergoes severe pathological alterations during the course of PD and may already be affected in the prodromal stages.33 PD pathology may lead to dysfunction and shrinkage of the amygdala and possibly with that the development of anxiety symptoms. There is some evidence from neuroimaging studies suggesting that amygdala volume is reduced in patients with PD compared with matched healthy controls34 ,35 and that (surgical) damage to the amygdala is associated with the development of anxiety.36 ,37 On the other hand, anxiety may itself cause of shrinkage of brain structures through overactivation of the hypothalamus-pituitary-adrenal axis (HPA) stress axis.38 Particularly the hippocampus seems very vulnerable to the effects of stress. Nevertheless, this mechanism appears less specific for the amygdala, that actually increases in volume due to dendritic hyperthrophy in the basolateral amygdaloid nucleus in response to stress-hormones.38 This increase in volume also correlated with the severity of anxiety symptoms in an animal model.39 The fact that the present study and studies in patients without PD who are anxious have generally reported decreases in amygdala volume13–16 suggests that factors other than overactivation of the HPA axis negatively influence amygdala volume in anxious patients. Considering our findings against the background of the above reviewed studies we tentatively hypothesise that the PD pathology is responsible for the observed volume loss of the left amygdala and the concomitant development of anxiety symptoms. The validity of this hypothesis needs to be investigated by future longitudinal studies on the directionality of PD development, amygdalar volume and anxiety symptoms.

Anxiety and depression frequently co-occur in PD and show overlap in symptoms and pathophysiology.19 In a previous VBM study we showed that the severity of depressive symptoms correlated negatively with volume in the bilateral hippocampus and right amygdala.18 The cluster we observed in the left hippocampus also extended somewhat into the left amygdala. In the present study the BDI was removed from the model during the stepwise selection procedure because it explained less variance in left amygdalar volume than the affective subscale. On the other hand, given the high correlation between anxiety and depressive symptoms we cannot fully exclude the possibility that the correlation is partly confounded by the severity of depressive symptoms. Indeed, if we correct the final model for the severity of depressive symptoms the association between the affective subscale and left amygdalar is no longer significant (data not shown). The exclusion of patients with comorbid depressive symptoms could have overcome this limitation. Nevertheless, we argue that such a group would not be representative of the majority of patients with PD with anxiety symptoms, in whom comorbidity with depression occurs frequently. Another possible limitation stems from the fact that we used the BAI questionnaire as a measure for anxiety symptoms, rather than a formal diagnosis of an anxiety disorder. A Movement Disorder Society task force classified the BAI and all other available anxiety rating scales as ‘suggested’ for PD rather than ‘recommended’, in part because they have limitations in their construct validity.1 We tried to overcome some of the shortcomings of the BAI by using an affective subscale of the BAI that, according to our previous factor analysis, is less susceptible to the severity of PD-related motor and autonomic symptoms.22 A recently developed anxiety rating scale for PD, the Parkinson Anxiety Scale (PAS), showed better clinimetric properties than other frequently used anxiety scales in PD.40 This rating scale can be used in future replication studies on the structural brain correlates of anxiety symptoms in PD. Lastly, because we did not include a group of healthy controls in our analyses we could also not disentangle the effects of the Parkinson pathology and naturally occurring variability in amygdala volume on the severity of anxiety symptoms.

In conclusion, using two different structural brain imaging analysis techniques we demonstrated that symptoms of anxiety in PD are associated with reduced volume of the left amygdala, independent of the severity of motor and autonomic symptoms and medication status. As this was a cross-sectional study, which impedes any clear-cut conclusions on the directionality of this association, we can only speculate that the PD pathology underlies shrinkage of the amygdala and constitutes a risk factor for the development of anxiety.


View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • YDVdW, OAvdH contributed equally.

  • Contributors CV took part in study design, analysis and interpretation of the data, writing, literature search. PSWB took part in analysis and interpretation of the data and revising manuscript. SR took part in study design, interpretation of the data, revising manuscript. HWB was involved in patient recruitment, study design, revising manuscript and supervision. OAvdH, YDvdW took part in study design, interpretation of the data, revising manuscript, study supervision and are guarantors.

  • Competing interests None declared.

  • Ethics approval Medical Ethical Committee VU University medical center.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.