Amygdala volume changes with posttraumatic stress disorder in a large case-controlled veteran group

MRI acquisition

Images were acquired on a General Electric 3 Tesla Signa EXCITE scanner equipped with an 8-channel headcoil. High-resolution T1-weighted whole brain images with 1-mm isotropic voxels using array spatial sensitivity encoding technique (ASSET) and fast spoiled gradient-recall (3D-FSPGR) were acquired axially for all participants. Image parameters were optimized for contrast between white matter, gray matter, and CSF (TR/TE/flip angle=7.484 ms/2.984 ms/12°, FOV=256 mm, 1-mm slice thickness, 166 slices, 256×256 matrix, 1 excitation).

Image analysis

All T1 images were visually inspected (CCH) to assure appropriate quality. Automated segmentation and labeling of the amygdala and hippocampus and estimation of total intracranial volume from participants’ T1 images were performed using the FreeSurfer image analysis suite³⁸ (version 5.0.0; http://surfer.nmr.mgh.harvard.edu/) and its library tool recon-all. We previously validated FreeSurfer automated segmentation of the amygdala and hippocampus compared to manual tracing²⁷. Spatial normalization by affine registration to Talairach space and skull stripping were performed on the T1 images. Registration was checked visually for accuracy (CCH). FreeSurfer segmentation and labeling of subcortical structures was based on a combination of voxel intensity, probabilistic atlas location, and the spatial relationships of voxels to the location of nearby subcortical structures. Using the FreeSurfer library function mri_label2vol and a transformation matrix generated by tkregister2, the segmentation labels were returned to native space. The native-space segmentations were converted to LAS orientation and then the amygdala and hippocampus were extracted using the segmentation labels. The segmentation of the amygdala and hippocampus overlayed on original T1 images (see Figure 1) was visually inspected (ALG, VMB, KSL) slice-by-slice for correct location and shape. All participants passed this inspection process without the need for manual adjustment.

Open in a separate window

Figure 1

Example of automated segmentation of the amygdala and hippocampus using FreeSurfer. The high-resolution structural MRI of a representative subject is displayed in radiological convention with segmentation labels for the amygdala (in blue) and hippocampus (in yellow) as shown in (A) sagittal (left hemisphere), (B) coronal, and (C) axial slices.

Volumetry results

Group means, standard deviations, and the GLM results testing our hypotheses on the effects of PTSD diagnosis, combat exposure, lifetime trauma, illness chronicity, and BDI are summarized in Table 3. The between-group results demonstrated that PTSD diagnosis was associated with smaller volume in the left and right amygdala and the left hippocampus. Right hippocampal volume was not significantly associated with PTSD diagnosis (see Table 3). Combat exposure was not significantly related to left amygdala, right amygdala, or right hippocampal volume, but showed a trend towards significance for left hippocampal volume. Lifetime trauma, illness chronicity, and depressive symptoms were not associated with volume differences in either structure for either hemisphere. Intracranial volume was significantly correlated with bilateral amygdala and hippocampus volumes (all p-values < .001). The other covariates including AUDIT, medication use, age, gender, BDI, CES, and PTSD duration were not significantly related to either amygdala or hippocampal volumes (data not shown; all p-values > .05). Results obtained by statistically controlling for medication effects were compared to a secondary analysis that excluded participants on antidepressant (n=28) and antipsychotic medications (n=7). Smaller volumes were confirmed in left amygdala [F(1,154)=8.67, p =. 004], right amygdala [F(1,154)=6.60, p =. 01], and left hippocampus [F(1,154)=5.74, p = .02]. The right hippocampus did not reach statistical significance [F(1,154)=1.89, p = .17].

Duration of PTSD was unavailable in eight PTSD participants, therefore analysis of the whole group omitted this covariate to include 99 participants with PTSD and 101 controls (see Table 4). The differences in volume between PTSD and non-PTSD groups in the right amygdala and left hippocampus were significant but had a higher p-value than with the inclusion of PTSD duration as a covariate, whereas the volume difference in the left amygdala was at trend level. The volume difference in the right hippocampus was again not significant. Severity of PTSD in patients assessed with the CAPS, controlling for ICV, was negatively correlated with volume of the left amygdala (r = −.22, p = .03), but did not reach significance in the right amygdala (r = −.18, p = .09). As expected, removing ICV as a covariate resulted in a non-significant correlation given the established association between head size and regional brain volume.

Data reduction

Two components, which individually explained 30.3% and 26.9% of the variance, respectively, had eigenvalues greater than 1. Selection of these two components was confirmed by parallel analysis⁴¹. The first component was associated with depressive symptoms, traumatic events, longer PTSD duration, older age, and combat exposure. The second component was associated with longer PTSD duration, younger age, alcohol abuse, and less combat exposure. A factor analysis that included intracranial volume showed it had low communality (.33), meaning the extracted components accounted for minimal ICV variance). Consequently, ICV was excluded from the final factor analysis. Results of GLM using covariates of components 1 and 2, intracranial volume, gender, use of serotonergic medication, and use of antipsychotic medication were similar to the original analyses conducted with all covariates. That is, PTSD diagnosis was associated with smaller left and right amygdala volume and left hippocampal volume (see Table 5).

Comparison to previous findings for amygdala volumetry

Although most previous studies reported non-significant findings, some studies showed a trend towards significance for smaller amygdala volumes in PTSD^{21,22,24,26,42}. The investigations of PTSD and amygdala volume with the largest samples included PTSD groups with n=44⁴³ and n=28⁴⁴, but were conducted in children and therefore do not generalize well to adults due to developmental changes in brain structure and connectivity⁴⁵. All studies of adults had a sample size of fewer than 20 in the PTSD group, with the majority of studies having 15 or fewer participants^20,22–26. Consistent with our findings and relevant to the core symptom cluster of re-experiencing symptoms in PTSD, smaller amygdala volume was associated with the presence of cancer-related intrusive recollections in a sample of 76 breast cancer survivors¹⁸.

It is worthwhile to consider the factors that may have produced a significant association of decreased amygdala volume given the preponderance of negative findings in prior studies. The small effect sizes observed from the meta-analyses suggest that they were underpowered to detect significant differences. Assessing a large sample size is impractical with manual segmentation, and this motivated our use of an automated segmentation technique. Automated segmentation also: (i) facilitates future replication of these results by other investigators; (ii) diminishes differences among studies due to the use of different protocols for defining anatomical boundaries; (iii) eliminates variability associated with different raters across studies and even a single rater over time; and (iv) removes bias introduced by varied software interfaces that are used in manual tracing (e.g., 3-plane views vs. single plane view). Thus, replication of the present results based on automated segmentation may be achieved with greater fidelity than manual tracing methods³⁸. Although we previously reported that FreeSurfer introduces additional variance compared to the gold standard of hand tracing^27,46, this variance would not bias volumetry measures in favor of one group over another.

Rather than matching individuals on a participant-by-participant basis to achieve the large enrollment necessary for this study in a realistic time frame, we used a statistical approach²¹ to control for potentially confounding variables (e.g., depressive symptoms, trauma load, duration of PTSD, intracranial volume, age, medication, alcohol abuse). The effects of many of these variables have not previously been tested as covariates in PTSD amygdala investigations. Assembling a PTSD group that is free of depressive symptoms is unlikely to generalize, and moreover new evidence calls into question whether PTSD and depression are distinct entities among individuals exposed to trauma⁴⁷. Notably, there was no significant association between depressive symptoms and amygdala or hippocampal volumes after controlling for PTSD among the other covariates. Larger amygdala volumes have been found in early depression⁴⁸ but smaller or null findings have been found in chronic depression⁴⁹. Nevertheless, it is important to exert statistical control for these symptoms. Exclusion of participants taking psychotropic medication has also been the accepted orthodoxy, although leaders in the field of PTSD neuroimaging have argued for their inclusion⁵⁰. Technical concerns in previous studies include the use of older MRI technology such as lower field strength scanners (1.5 or 2.0 Tesla, see Table 1)^{19,21–24,42–44} and lower spatial resolution (≥ 1.5 mm slice thickness)^{19–26,42–44}. We addressed these issues with a 3 Tesla scanner to improve SNR and 1-mm slice thickness (1mm isometric) for superior spatial resolution. Based upon our previously reported power calculation for amygdala segmentation with FreeSurfer, a sample size of at least 55 subjects per group is required for the effect sizes we observed²⁷.

In addition to these methodological improvements, important demographic and trauma-related characteristics of our sample differ from those of previous studies. In contrast to some earlier studies of veterans with chronic PTSD^23–24,51, our participants were recent military personnel who served largely in military conflicts in Iraq and Afghanistan.

Associations with functional neuroimaging of amygdala

Functional differences of the amygdala, particularly in the left hemisphere, during emotion processing were supported by a meta-analysis showing ventral anterior hyperactivation and a dorsal posterior hypoactivation in PTSD⁹. Amygdala engagement during fear conditioning is well established in healthy adults. Thus, amygdala hyperactivity in PTSD may reflect an exaggerated response of fear circuitry and may explain PTSD symptoms such as hypervigilance and hyperarousal. Despite concerns with statistical power, heterogeneity of task design, patient characteristics, imaging modality, and analytic approaches in functional neuroimaging studies of PTSD, these results have been more consistent than results from volumetric studies of the amygdala. Overall, there are numerous reports of greater amygdala activation in PTSD^{6–7,52–55}, whereas some others failed to show increased amygdala activation^56–58, further obfuscating a coherent hypothesis for amygdala volume differences in PTSD. Indeed, amygdala volume decrease appears to correspond to increased fMRI functional activation in PTSD⁵⁹. The ventral hyperactive cluster reported by Etkin and Wager⁹ may relate to the basolateral amygdala (described below) and may be relevant to acquired fear responses in PTSD given the role of this region in forming emotional memories^60–61.

Pathobiology of PTSD

The amygdala plays a key role in a wide variety of behaviors and mnemonic functions, most critically in modulating negative affect and emotion.^{3, 63, 64} Changes in fear, stress, and anxiety in rats are induced by lesions to subnuclei of the amygdala, notably the BLA, which is needed to form and later express associative fear memories. Evidence primarily from rodent work has led to a model in which the lateral nucleus in the BLA receives aversive and sensory signals that are passed on to the basal nucleus and central nucleus. Fear-associated behaviors are governed by the central nucleus, which provides the major outputs of the amygdala.^65,66 This model is consistent with the observation that prior to fear conditioning, rats with lesions to the lateral nucleus did not form fear memories, unlike lesions to the basal nucleus.^67,68 However, following fear conditioning, lesions to the basal nucleus blocked the expression of fear memories, but not the ability to encode those memories^.69 Thus, fear-related plasticity in the amygdala is essential for fear learning and accompanying fear behaviors.^11,66

The effect of persistent and chronic threat-induced hyperexcitation of the amygdala on its volume in humans is yet unclear. Much of our knowledge on the pathobiology of threat and stress effects on the brain, and on the medial temporal lobe in particular, comes from animal models using acute stressors that are qualitatively different from the chronic and/or extreme stressors typically experienced by humans with PTSD. One of several competing theories is based on established findings that the amygdala and hippocampus undergo stress-induced structural remodeling, albeit in very different ways.¹⁰ Most of the research on stress effects has focused on the hippocampus, with very few studies investigating the amygdala directly.

Smaller amygdala volume: vulnerability or consequence?

The second goal of this study was to gain insight into whether the smaller amygdala volume is a preexisting vulnerability factor for developing PTSD or a consequence of having PTSD. Our data failed to show a correlation of trauma load or PTSD chronicity with lower amygdala volume, suggesting the lack of a dose-response effect for trauma and amygdala volume or that smaller amygdalae might be a risk factor for developing PTSD.

Evidence from prior work in humans and animals demonstrates an association between smaller amygdalae and stronger fear conditioning and stress reactivity, which are considered risk factors for PTSD. Prior work in animals by Yang and colleagues¹¹ used recombinant inbred strains of mice that exhibit up to a 2-fold difference in BLA size.¹¹ Mice were categorized into groups with small, medium, and large BLA volume and underwent a pavlovian fear-conditioning procedure. The small-BLA mice showed stronger fear conditioning than the medium- and large- BLA mice, and freezing to the conditioned stimulus was significantly correlated with volume of the BLA but not of comparison regions, including the hippocampus, striatum, or cerebellum. Mice were also subjected to a stress condition (forced swim), which was associated with elevated corticosterone level in the small-BLA but not the medium- or large-BLA groups. Nonstressed mice did not differ by corticosterone level. The BLA is a critical site through which corticosterone enhances associative fear memories.

The BLA volume association raises a fundamental question whether a small BLA is the consequence or the cause of stronger fear conditioning. A smaller BLA is unlikely to be a consequence given that chronic threat and stress lead to corticosterone-mediated spinogenesis and dendritic arborization in mice.^{10, 70–72} In clear contradistinction, the hippocampus shows loss of neurons and synaptic connectivity in response to elevated adrenal hormone levels associated with chronic threat and stress.^73–75 Thus, the smaller volume of the BLA, which constitutes the largest of the amygdalar nuclei, is linked to stronger fear conditioning, chronic threat, and vigilance, leading to a potential increase in BLA volume.⁵ Although the effects of chronic threat and stress in rodents may differ in humans, the neural systems for fear processing are highly conserved across phylogeny. The evidence of counteracting influences on amygdala volume found in rodents is likely to cloud any association of amygdala volumetry and PTSD and may explain the inconsistent and sometimes conflicting reports of amygdala volumes in PTSD.^{14, 15}

Evidence from 2 studies in humans builds on the work of Yang and colleagues¹¹ in mice. Studies in humans not exposed to trauma or chronic threat and stress are especially informative because trauma-induced structural changes in the amygdala are unlikely to have occurred. First, Hartley et al ¹³ paired colored squares (conditioned stimuli) with mild electrical shock (unconditioned stimuli) and measured the strength of fear acquisition via skin conductance response. The magnitude of the conditioned fear response was correlated with the smaller amygdala volume. Second, Gianaros et al¹² measured mean arterial pressure in response to a stressor (performance-titrated Stroop Color-Word Interference task) and found that a smaller amygdala volume was correlated with stressor-evoked blood pressure reactivity.

To investigate directly whether our finding of decreased amygdala volume in PTSD represents a preexisting vulnerability factor or an acquired sign of the disorder, research studies using prospective, longitudinal design and twin-discordance models are needed.⁷⁶ Vulnerability related to genetic or epigenetic effects has been hotly debated, but little empirical evidence is available. Altered serotonin binding in the left amygdala⁸ and increased left amygdala activation modulation by the 5HTTLPR serotonin transporter gene⁷⁷ have been observed in PTSD. However, evidence showing specific genetic modulation of amygdala volume in PTSD is lacking.

Limitations

We used covariates to control for variables such as depressive symptoms, alcohol abuse, and age. This approach is adequate unless a nonlinear association is present. For instance, trauma exposure may not follow a linear dose response but instead may require a specific threshold beyond which a marked effect on amygdala volume is produced. Secondary analyses that matched groups for trauma exposure were performed to rule out this possibility. Also, our sample consisted of veterans from the Iraq and Afghanistan conflicts, who were mostly men; therefore, owing to these and other sources of selection bias, we urge caution when generalizing these results to other demographic groups. Our sample also showed high levels of combat trauma relative to other trauma types, and it remains unclear whether the type of trauma predicts the magnitude of volume loss in the amygdala. Approximately one-fourth of the sample was diagnosed using the DTS, a self-report measure that has high predictive power compared with the CAPS but may misclassify some subjects. Finally, based on the effect sizes we obtained for the hippocampus, the present sample size may have been insufficient to detect volume differences in the right hippocampus; however, this effect size was consistent with a large meta-analysis.¹⁵

This research was supported by the National Institutes of Health Grant K23 MH073091 (RAM), the Department of Veterans Health Affairs (VHA), GRANT00507153, GRANT10490938, and Mental Illness Research Education and Clinical Center Grant for Post-Deployment Mental Health.