The brain at war: effects of stress on brain structure in soldiers deployed to a war zone

Stress is ubiquitous phenomenon in our daily lives and its negative impact on mental and physical health has long been recognized. However, effects of extremely stressful life events such as the experience of acts of terrorism, natural disaster or military combat are difficult to study, since these events are rare and less predictable. Individuals respond very differently to traumatizing events, with some showing resilience and others developing psychiatric diseases such as post-traumatic stress disorder (PTSD) or depression. Cross-sectional neuroimaging studies in PTSD have frequently reported structural deficits in the hippocampus and medial prefrontal cortex, including anterior cingulate cortex (ACC) and ventromedial prefrontal cortex (vmPFC), in patients compared to trauma-exposed controls. However, cross-sectional and many prospective designs cannot clarify causality: whether brain structural alterations are pre-existing vulnerability factors or consequences of trauma exposure/disease. To address causality, monozygotic twin designs have been used, comparing twins discordant for PTSD and for trauma exposure, leveraging shared genetics and upbringing to infer vulnerability versus acquired effects. The present longitudinal study investigates brain structural changes in soldiers deployed to a war zone versus non-deployed controls to disentangle effects of trauma exposure on brain structure.

Prior cross-sectional studies consistently implicate reduced hippocampal volume and structural alterations in medial prefrontal regions (ACC, vmPFC) in PTSD compared to trauma-exposed controls. These designs, however, cannot determine if observed differences are pre-existing risks or acquired changes. Monozygotic twin studies, including pairs discordant for PTSD and for trauma exposure, attempt to parse vulnerability from consequence by controlling for shared genes and early environment. Meta-analytic coordinates used in this study’s ROI selection derive from literature contrasting PTSD patients to trauma-exposed controls, highlighting ACC (0, 24, 8), vmPFC (0, 49, 6), and hippocampus (−30, −14, −14) as regions of interest.

Design: Longitudinal cohort with deployed (combat) and non-deployed control soldiers. MRI assessments at Baseline and Follow-up I for both groups; a third Follow-up II for the combat group only. Participants: n = 121 combat; n = 40 controls. Baseline PTSD scores (PDS) were below clinical cut-offs (28 per original; 24 per German sample). Intervals: Combat group Baseline to Follow-up I mean 216 days (SD = 68); Controls Baseline to Follow-up I mean 214 days (SD = 85); no group difference (t(158) = 0.13, p = 0.90). Combat group Follow-up I to Follow-up II mean 189 days (SD = 107). MRI acquisition: Siemens Tim Trio 3T, standard 2-channel head coil. T1-weighted MPRAGE: TR = 2500 ms; TE = 4.77 ms; TI = 1100 ms; matrix 256 × 256 × 176; flip angle 7°; 1 mm isotropic voxels. Image processing: CAT12 v1.78 on SPM12 (Matlab R2016a); longitudinal pipeline with intra-subject realignment, bias correction, segmentation, normalization estimated on mean image across timepoints and applied to all images; smoothing 6 mm FWHM. Visual artifact checks pre- and post-segmentation; statistical quality control based on inter-subject homogeneity. Behavioral measures: Self-report questionnaires at time points included PDS (PTSD symptoms), STAI-State, ASI, BDI-II, RSQ, AUDIT (German). Combat Experience Scale (CES) administered before and after deployment to index exposure. Regions of interest: Meta-analysis–derived ROIs: ACC (MNI 0, 24, 8), vmPFC (0, 49, 6), left hippocampus (−30, −14, −14); ROI extraction via MarsBaR; additional anatomical masks based on AAL for amygdala analyses. Statistical analysis: Group × time repeated-measures analyses on questionnaires, controlling for age (group age difference present at baseline). For BDI, non-parametric robust RM-ANOVA with 10% trimmed mean. Bayesian statistics (JASP) used for hippocampus interaction to assess evidence for null. VBM group × time interaction analysis focused on within-subject changes; covariates included age at MRI; sex and total intracranial volume were not included. Thresholding: voxel-level p < 0.001; cluster-level FWE correction p < 0.05 with non-stationary smoothness correction. Effect sizes reported as partial eta squared (η²p). Post-hoc paired t-tests examined within-group changes across intervals; multiple-comparisons correction applied to post-hoc tests.

Questionnaires: No significant group × time differences for PDS (PTSD symptoms), STAI-State, ASI, RSQ, or AUDIT. Significant group × time interaction for BDI-II (parametric: F(1,158) = 5.52, p = 0.023, η² = 0.032; non-parametric: F(1,108.33) = 63.86, p < 0.001), reflecting a small increase in depressive symptoms in the combat group (Δ ≈ 0.76; t(120) = −2.60, p = 0.010) and a non-significant numerical decrease in controls (t(39) = 1.15, p = 0.256). Absolute BDI levels remained minimal (≈2.6–3.4). ROI VBM analyses: Group × time interactions observed for ACC (F(1,158) = 5.10, p = 0.025, η² = 0.031) and vmPFC (F(1,158) = 5.29, p = 0.023, η² = 0.032), but these did not survive Bonferroni correction for three ROIs (p < 0.0167). No interaction in hippocampus (F(1,158) = 1.62, p = 0.211). Bayesian mixed ANOVA for hippocampus showed substantial evidence favoring the absence of interaction (BF01 = 2.41; BF01 with interaction = 2.689 vs without interaction = 1.115). Post-hoc within-combat-group tests showed significant gray matter decreases: ACC Baseline→Follow-up I t(120) = 3.90, p < 0.001; Baseline→Follow-up II t(91) = 10.24, p < 0.001; Follow-up I→Follow-up II t(91) = 8.69, p < 0.001. vmPFC showed a similar pattern: Baseline→Follow-up I t(120) = 4.43, p < 0.001; Baseline→Follow-up II t(91) = 8.14, p < 0.001; Follow-up I→Follow-up II t(91) = 4.90, p < 0.001. Controls showed no significant ACC or vmPFC change Baseline→Follow-up I (ACC t(39) = −0.48, p = 0.634; vmPFC t(39) = −0.38, p = 0.709). Amygdala ROIs: no significant interactions (left: F(1,158) = 0.27, p = 0.607; right: F(1,158) = 1.26, p = 0.263). Whole-brain VBM: Significant group × time interaction cluster in bilateral thalamus (peak MNI −9, −30, 4; 712 voxels; FWE-corrected). A baseline group difference was present (controls > combat; t(120) = −3.27, p = 0.001). Within combat group, significant decrease Baseline→Follow-up I (t(120) = 5.59, p < 0.001), with no additional significant change Follow-up I→Follow-up II (t(91) = 1.27, p = 0.21). Post-hoc tests survived multiple-comparisons correction. Volumetric reductions did not correlate with increases in PTSD symptoms. Sex distribution did not differ by group (χ² = 0.97, p = 0.325).

This longitudinal study aimed to identify structural brain changes attributable to deployment-related stress exposure. Despite many deployed soldiers reporting combat-related stressors, there were no meaningful increases in PTSD symptoms, anxiety, anxiety sensitivity, rumination, or alcohol use, and only a small rise in depressive symptoms within the minimal range. Imaging revealed gray matter volume decreases in medial prefrontal regions (ACC and vmPFC) and bilateral thalamus in the deployed group. mPFC reductions progressed beyond the deployment period, continuing over approximately six months post-deployment, whereas thalamic reductions were evident by post-deployment without further decline by the second follow-up. The absence of hippocampal effects and Bayesian support for the null interaction suggest that hippocampal volume loss is not a necessary consequence of stress exposure in this context. The pattern that changes occur in deployed soldiers and not in controls supports the interpretation that these volumetric reductions reflect acquired effects of trauma/stress exposure rather than pre-existing vulnerability markers. The continued decline in mPFC volume after return from deployment may indicate an ongoing pathophysiological process below the threshold of overt clinical symptoms. These findings refine the neurobiological model of stress-related brain changes, emphasizing mPFC and thalamus as sensitive to real-world combat exposure even when clinical symptom measures remain low.

Using a prospective longitudinal design in active-duty soldiers, the study demonstrates gray matter volume reductions associated with deployment-related stress in the ACC/vmPFC and bilateral thalamus. mPFC changes continued for months after deployment, while thalamic decreases appeared to stabilize. No hippocampal interaction effects were found, with Bayesian analyses supporting the null. Volumetric changes did not track increases in PTSD symptoms, suggesting they reflect exposure-related brain alterations rather than vulnerability markers or symptom severity. Future research should extend follow-up durations, incorporate detailed and validated exposure metrics, examine biological mechanisms underlying continued mPFC decline, include clinical PTSD cases to assess symptom–structure coupling, control for additional covariates (e.g., TIV) and potential confounds, and test interventions that might mitigate or reverse exposure-related structural changes.

1) Exposure measurement issues: CES difference scores yielded implausible decreases in 26.5% of soldiers, complicating quantification of deployment exposure. 2) Multiple comparisons: ACC and vmPFC group × time interactions did not survive Bonferroni correction across the three ROIs, tempering confidence in ROI-level inferences. 3) Baseline group difference in thalamic gray matter (controls > combat) complicates interpretation of interaction effects. 4) Covariates: Analyses focused on within-subject change and controlled for age, but not sex or total intracranial volume, which could influence volumetric estimates. 5) Symptom range: Low symptom levels and minimal changes (e.g., BDI in minimal range) may limit detection of structure–symptom relationships. 6) Generalizability: Sample comprised active-duty soldiers with subclinical PTSD symptoms; findings may not generalize to civilians or clinically diagnosed PTSD populations. 7) Follow-up asymmetry: Only the combat group had a second follow-up, limiting comparative longitudinal inferences. 8) Self-report measures and potential recall/reporting biases in exposure and symptoms.