Differential relationships of PTSD symptom clusters with cortical thickness and grey matter volumes among women with PTSD

Structural neuroimaging studies of posttraumatic stress disorder (PTSD) have typically reported reduced cortical thickness (CT) and gray matter volume (GMV) in subcortical structures and networks involved in memory retrieval, emotional processing and regulation, and fear acquisition and extinction. Although PTSD is more common in women, and interpersonal violence (IPV) exposure is a more potent risk factor for developing PTSD relative to other forms of trauma, most of the existing literature examined combat-exposed men with PTSD. Vertex-wise CT and subcortical GMV analyses were conducted to examine potential differences in a large, well-characterized sample of women with PTSD stemming from IPV-exposure (n = 99) compared to healthy trauma-free women without a diagnosis of PTSD (n = 22). Subgroup analyses were also conducted to determine whether symptom severity within specific PTSD symptom clusters (e.g., re-experiencing, active avoidance, hyperarousal) predict CT and GMV after controlling for comorbid depression and anxiety. Results indicated that a diagnosis of PTSD in women with IPV-exposure did not significantly predict differences in CT across the cortex or GMV in the amygdala or hippocampus compared to healthy controls. However, within the PTSD group, greater re-experiencing symptom severity was associated with decreased CT in the left inferior and middle temporal gyrus, and decreased CT in the right parahippocampal and medial temporal gyrus. In contrast, greater active avoidance symptom severity was associated with greater CT in the left lateral fissure, postcentral gyrus, and middle/lateral occipital cortex, and greater CT in the right paracentral, posterior cingulate, and superior occipital gyrus. In terms of GMV, greater hyperarousal symptom severity was associated with reduced left amygdala GMV, while greater active avoidance symptom severity was associated with greater right amygdala GMV. These findings suggest that structural brain alterations among women with IPV-related PTSD may be driven by symptom severity within specific symptom clusters and that PTSD symptom clusters may have a differential (increased or decreased) association with brain structures.

ducted at two sites: the University of Arkansas for Medical Sciences (UAMS) and the University of Wisconsin-Madison (UW). All study procedures were approved by the Institutional Review Board at UAMS and the UW Health Sciences Institutional Review Board, and all methods were carried out in accordance with relevant guidelines and regulations. The analyses included in the current manuscript are independent of previously reported findings [52][53][54][55] . These studies recruited women with a primary diagnosis of PTSD and a primary index trauma of IPV-exposure, along with healthy women, free of trauma exposure or mental health concerns. Inclusion criteria for the PTSD group in all studies was a primary diagnosis of PTSD, IPV-exposure, and female sex. Inclusion criteria for the healthy control groups included female sex, absence of trauma exposure, absence of current psychiatric disorders, and absence of psychiatric medication use. Exclusion criteria for all participants included internal metal or other MRI contraindications, major medical disorders, and endorsement of psychotic symptoms.
Assessments. Participants completed a baseline demographic questionnaire assessing age, race, and education. Trauma histories were assessed with the National Women's Survey (NSA) 56 trauma assessment section, which is a structured interview that assesses previous history of physical abuse by a caregiver, physical assault, sexual assault, witnessed domestic violence, witnessed community violence, and a range of other stressful life events. The presence of mental health and psychiatric disorders (e.g., major depressive disorder, anxiety disorders) was assessed using the Structured Clinical Interview for DSM-IV 57 . PTSD diagnosis and symptom severity was assessed using the Clinician Administered PTSD Scale for DSM-5 (CAPS-5) 58 and the PTSD Checklist for DSM-5 (PCL-5) 59 for two studies (n = 110) and the Structured Clinical Interview for DSM-IV Disorders (SCID-IV) and PTSD Checklist-Civilian Version (PCL-C) 60 for one study (n = 30). Of the 121 participants included in the analyses, 11 participants (50%) from the control group and 15 participants (15%) from the PTSD group completed DSM-IV measures, while 11 participants (50%) from the control group and 84 participants (85%) from the PTSD group completed DSM-V measures. PCL-5 scores were used as the measure of PTSD symptom severity for each cluster and for overall symptom severity in all analyses including PTSD symptoms. As a result, PCL-C scores were converted to PCL-5 scores using an established and validated crosswalk procedure using equipercentile equating 61  Quality control and attrition. All T1-weighted images were visually inspected to identify and exclude poor quality images. Following inspection, 17 images were excluded due to poor quality T1s, due to participant head movement (n = 16), and a neuroanatomical anomaly (n = 1). Additionally, two individuals had high quality T1 images, but were missing PCL-5 scores. The remaining 121 raw T1-images were retained and consisted of the final sample for data analyses.
Image preprocessing. Preprocessing of the raw T1-weighted images was completed in FreeSurfer Version 6.0.0 (http://surfe r.nmr.mgh.harva rd.edu) 62 on a Linux platform. FreeSurfer's reconall command was used to preprocess all images through standard steps, including motion correction and averaging 63 , removal of nonbrain tissue and skull-stripping 64 , automated Talariach transformation, segmentation of subcortical white and gray matter 62,63 , transformation to MNI305 atlas space, intensity normalization 65 , and volumetric registration. Technical details for FreeSurfer are described in detail elsewhere 62-64,66-72 . Image analysis. For all images, reconall continued without intervention through the processing stream following initial preprocessing steps. This included steps for another intensity normalization, white matter segmentation and editing, tessellation of white and gray matter boundaries and automated topology correction 73,68 , smoothing and surface inflation, spherical mapping and registration for each hemisphere 67,68,70 , and parcellation and labelling of cortical and subcortical structures. Cortical structures were labelled according to the Desikan-Killiany-Tourville parcellation (DKT) 74 and mapped to the spherical surface, while subcortical and white matter structures are labelled according to FreeSurfer's automated segmentation tool (aseg) 58 and mapped to the MNI305 common space. Final statistics for GMV, CT, and surface area estimates of each region by hemisphere in both the DKT and subcortical aseg parcellations were generated.
Vertex-wise cortical thickness analysis. Two separate vertex-wise CT analyses were conducted to: (1) examine group differences between control and PTSD groups; and (2) examine whether PTSD symptom severity scores for each PTSD symptom cluster are associated with CT in the PTSD group only. To conduct these vertex-wise analyses of potential differences in CT across the brain, we first generated cortical surface files for each subject. Subject surface files were registered to the average template (fsaverage), which is constructed from scans of 40 healthy adult subjects 75  www.nature.com/scientificreports/ conducted using either group (control vs PTSD) or PTSD symptom severity scores for PTSD symptom clusters (i.e., re-experiencing, active avoidance, negative alterations in cognition and mood, and hyperarousal) as the main predictor variables for each separate set of analyses (i.e., group [PTSD vs control group] analysis, and PTSD symptom cluster analysis). Additionally, given that the control group consisted of non-trauma exposed women, exploratory analyses (see supplementary material) were conducted examining trauma exposure (i.e., number of IPV-related direct assaults experienced) as a main predictor of interest (as opposed to group: control vs PTSD). All comparisons were conducted using AFNI's 3dttest + + function 76 , controlling for participant age, image acquisition site (UW or UAMS), and current diagnoses of major depression or anxiety disorders. To control for multiple comparisons, we used the AFNI slow_surf_clustsim.py function to run clustsim on the grouplevel fsaverage surface. We specified the uncorrected p threshold at 0.001 and ran 10,000 Monte Carlo simulations to generate the minimum surface cluster area unlikely to be detected due to chance (p > 0.05). Finally, cortical regions with thickness that significantly related to either group or PTSD symptom severity for each PTSD symptom cluster above the area threshold were extracted using AFNI SurfClust.
Gray matter region of interest selection. Gray matter regions of interest (ROIs) included the hippocampus and amygdala of both the left and right hemispheres. ROIs were selected a priori based on previous research suggesting reduced volume in PTSD as a result of various trauma exposures 22-24 . Gray matter data analysis. Following processing in Freesurfer, gray matter ROIs from the DKT parcellation were entered into multiple linear regression models. Two sets of models were tested for each hemisphere for both ROIs (amygdala and hippocampus). Each model started off including covariates of non-interest (age, site, estimated total intercranial volume [eTIV]), before adding the main predictors of interest (i.e., group, PTSD cluster symptom severity scores, respectively for each set of models), followed by the inclusion of additional covariates (i.e., current diagnosis of anxiety disorder and major depressive disorder, coded as separate variables) that are prevalent within PTSD, and therefore may explain unique variance. The model examining group differences in GMV used dummy-coded contrasts with the control group as the reference group. Models were estimated using the fitlm function in Matlab. In order to avoid an inflated Type 1 error rate, we corrected for multiple comparisons by dividing the p-value of 0.05 by 4 (p = 0.0125). Lastly, given that the control group consisted of non-trauma exposed women, exploratory analyses (see supplementary material) were conducted examining trauma exposure (i.e., number of IPV-related direct assaults experienced) as a main predictor of interest (as opposed to group: control vs PTSD). Table 1   for whole brain comparison) predict differences in CT across the entire cortical surface after controlling for age, scanner site, anxiety, and depression.

Participant characteristics. See
PTSD symptom clusters. Within the PTSD group, greater re-experiencing symptoms were significantly associated with decreased CT in the left inferior and middle temporal gyrus (Fig. 1A,B) and decreased CT in the right parahippocampal and medial temporal gyrus (Fig. 1C,D). Greater active avoidance symptoms were significantly associated with greater CT in the left lateral fissure ( Fig. 2A,B), postcentral gyrus (Fig. 2C,D), and middle/lateral occipital gyrus (Fig. 2E,F) and greater CT in the right paracentral (Fig. 3A,B), posterior cingulate, and superior occipital gyrus (Fig. 3C,D). Symptoms of hyperarousal and negative alterations in cognition and mood did not significantly (p > 0.05 corrected for whole brain comparison) predict differences in CT across the entire cortical surface after controlling for age, scanner site, anxiety, and depression.

Volumetric analyses. Control versus PTSD.
A diagnosis of PTSD did not significantly predict differences in amygdala (left and right, ps = 0.739 and 0.636, respectively) or hippocampal (left and right, ps = 0.696 and 0.337, respectively) volume (see Table 2; Fig. 4A,B, respectively). There was a trend toward a current diagnosis of major depressive disorder predicting less right amygdala volume (p = 0.020; see Table 2).
PTSD symptom clusters. Within the PTSD group, greater hyperarousal symptoms were significantly (p = 0.007) associated with less left amygdala volume, even after controlling for anxiety and depression (see Table 3; Fig. 4C).
Greater active avoidance symptoms were significantly (p = 0.012) associated with greater right amygdala volume, even after controlling for depression and anxiety (see Table 3; Fig. 4D). Re-experiencing and negative alterations in cognition and mood symptoms did not significantly predict left (ps = 0.923 and 0.757, respectively) or right (ps = 0.424 and 0.803, respectively) amygdala volume. PTSD symptom cluster scores did not significantly predict left (ps = 0.134 to 0.647) or right (ps = 0.338 to 0.996) hippocampal volume (see Table 3).

Discussion
This study investigated CT and subcortical GMV among adult women with PTSD resulting from IPV-exposure and a group of healthy, age-matched women free of trauma exposure or mental health concerns. Results from the current study revealed that a diagnosis of PTSD among IPV-exposed women did not significantly predict differences in CT across the cortex or GMV in the amygdala or hippocampus compared to trauma-free women without PTSD. However, re-experiencing, active avoidance, and hyperarousal symptom severity within the PTSD group were significantly associated with differential (i.e., increased and decreased) CT in several regions and GMV in the amygdala. Previous research among adults with PTSD, mostly male veterans, has consistently demonstrated reduced CT in several regions including the anterior cingulate cortex, prefrontal cortex (medial, ventral, dorsolateral), and inferior and superior temporal gyri 17,19,20,23,77 . Additionally, reductions in subcortical GMV within the hippocampus and amygdala compared to age-matched, healthy controls have also been consistently reported within this population 25,26 . Our findings, obtained from a vastly different population of women with PTSD following IPV-exposure are in contrast with these reports, as we found no differences in CT or GMV when group was dichotomized (PTSD vs control). Interestingly, our findings are consistent with a smaller (but more relevant) literature specifically focusing on homogenous samples of women with PTSD due to physical and sexual abuse. Consistent with our findings, Landre et al. examined a small sample of adult women with PTSD stemming from sexual abuse and found no differences in CT compared to controls 37 . Relatedly, there were no differences in hippocampal volume between women survivors of intimate partner violence and healthy controls 38 . Most recently, it was reported that adolescents with childhood sexual abuse exhibit preserved CT in the ventromedial prefrontal cortex, anterior cingulate gyri, middle temporal gyri, and superior temporal gyri compared to traumafree controls 40 . Collectively, there is growing evidence suggesting that the specific type of trauma exposure (e.g., combat vs. IPV) may explain whether a diagnosis of PTSD predicts alterations in CT and GMV compared to healthy, trauma-free controls.
Although a diagnosis of PTSD did not predict differences compared to healthy controls, specific clusters of PTSD symptoms were associated with CT and GMV within women with IPV-related PTSD. These results suggest that altered brain structure within IPV-exposed women with PTSD may be driven by symptom severity within specific clusters (e.g., re-experiencing, avoidance, hyperarousal), or that these particular alterations led to an increased likelihood of the emergence of more severe symptoms. Moreover, our results revealed that the symptom clusters had a differential effect on brain structure. For instance, greater active avoidance symptoms were associated with greater CT in several regions and greater GMV in the right amygdala, whereas greater reexperiencing and hyperarousal symptoms were associated with decreased CT in several regions and decreased left amygdala volume. The observed differential relationships with the amygdala depending on PTSD symptom clusters is consistent with a recent meta-analysis that found among adults with PTSD, greater re-experiencing symptom severity scores were associated with greater left amygdala activity in response to threat, whereas greater www.nature.com/scientificreports/ avoidance symptom severity scores were associated with less amygdala activity in response to threat 78 . Moreover, a prior investigation reported a positive relationship between avoidance symptoms and right amygdala (and hippocampus) volume, and an inverse relationship between hyperarousal symptoms and left superior medial frontal gyrus volume in a small sample of adults (66% women) with non-combat related PTSD (87% of sample experienced IPV-exposure) 39 . The differential relationships of PTSD symptom clusters with CT and GMV, in particular regions canonically implicated in PTSD (i.e., amygdala), highlights the importance of acknowledging that 'PTSD' is a diagnostic construct with considerable symptom heterogeneity 79 . Accordingly, researchers may be missing important structural correlates of PTSD in this specific population (women with IPV-exposure) and others (e.g., combat, motor vehicle accidents) by neglecting to examine the effect of each PTSD symptom cluster. For instance, our results revealed decreased CT in the parahippocampal gyrus of adults with greater re-experiencing symptoms. The parrahippocampal gyrus is a region of the cortex that surrounds the hippocampus and plays a role in encoding, and episodic, spatial, and contextual memory. Relatedly, re-experiencing symptoms involve memory retrieval processes. Although investigations are limited, there are two additional reports of decreased parahippocampal volume in PTSD populations (male combat veterans and natural disaster survivors) compared to controls 21,23 . Our findings add to a growing body of evidence suggesting that decreased CT in the parahippocampal gyrus may be a commonly observed structural alteration in PTSD, especially among those with greater re-experiencing symptoms. In contrast, greater avoidance symptoms were associated with greater CT in several regions (left lateral fissure and post central gyrus; right paracentral, posterior cingulate, and superior occipital cortex). Given that CT in these regions did not differ between controls and those with PTSD, further research is warranted to determine whether the observed structural alterations among women with PTSD with greater avoidance symptoms is clinically relevant. Lastly, it is worth noting that the addition of major depressive disorder (MDD) into our www.nature.com/scientificreports/ models yielded some intriguing patterns. For instance, there was an effect for depression to predict left and right amygdala volume in the between group analyses. Though it did not survive correction for multiple comparison, this may suggest that a current diagnosis of MDD, as opposed to PTSD, may be more predictive of amygdala volume in adult women with PTSD. Within the PTSD group, there was also a trend for depression to predict right amygdala volume. Although these findings were not statistically significant after controlling for multiple comparisons, and therefore their inclusion in the models did not influence the interpretation of the PTSD subgroup analyses, they highlight the potential importance of examining mental health comorbidities in addition to PTSD. For instance, Bremner et al. previously reported reduced hippocampal volume in women with PTSD from childhood sexual abuse and a comorbid diagnosis of MDD compared to individuals with current MDD without abuse, but only after controlling for PTSD 80 . While it is plausible that comorbid depression and anxiety may explain unique variance in predicting GMV, that does not appear to be the case for CT, as neither were associated with CT across the cortex in all tested models. As was done in the current study, it may be beneficial Table 2. PTSD vs control group as predictor of hippocampus and amygdala gray matter volume. A diagnosis of PTSD did not significantly (p > .0125) predict differences in amygdala (left and right) and hippocampal (left and right) volume. TIV = total intracranial volume.

Predictors t-stat P value t-stat P value t-stat P value t-stat P value
Left www.nature.com/scientificreports/ for future research investigations examining PTSD populations to present results with and without comorbid depression and anxiety disorder diagnoses included in statistical models in order to improve our understanding of each diagnoses unique contributions. This study is notwithstanding limitations. For instance, the sample size of the control group (n = 22) was much smaller than that of the PTSD group (n = 99), which may potentially explain the lack of group differences in CT and GMV. However, the control group sample size is relatively greater than many of the previous neuroimaging studies that have yielded similar findings 36,37 , and the sample size of the PTSD group (n = 99) and overall sample (n = 121) affords greater statistical power than prior studies. Additionally, the control group consisted solely of non-trauma exposed adults, which conflates exposure to trauma with the manifestation of PTSD (for our PTSD group). However, exploratory analyses incorporating trauma exposure as a main predictor yielded similar findings as results obtained from including a dichotomous group classification (i.e., control vs. PTSD). Another potential limitation is that anxiety and depression were simply dichotomized based on whether participants had a current diagnosis (i.e., endorsed diagnostically significant symptoms). The inclusion of depressive and anxiety symptoms on a continuum may have explained more variance for those who were experiencing subthreshold symptoms. Additionally, this was a cross-sectional study, rendering it impossible to determine whether these structural alterations (or lack thereof) were present prior to, or a direct consequence of IPV-exposure and/or the development of PTSD. Moreover, over 80% of the sample was Caucasian. Although there were no significant differences in race between groups, this highlights the need for additional research with a larger and more racially diverse sample. Lastly, the current study did not track hormonal variations among all participants during the time of their scan, which is a limitation as recent various hormonal factors (e.g., birth control, menstrual cycle stage; sex hormones) are known to influence gray matter volume [81][82][83][84] . However, exploratory analyses among the PTSD group (for which data was available) revealed that results examining symptom severity within specific PTSD symptom clusters as a predictor of amygdala and hippocampal GMV did not differ based on whether birth control use was included as a covariate (see Supplementary Table 3).  A,B). However, within the PTSD group, greater hyperarousal symptoms significantly predicted less left amygdala volume, even after controlling for anxiety and depression (C); whereas greater active avoidance symptoms significantly predicted greater right amygdala volume, even after controlling for anxiety and depression (D). Hyperarousal (cluster E; score range of 0-24) and active avoidance (cluster C; score range of 0 to 8) symptom severity scores were obtained from the PTSD Checklist for DSM-V (PCL-5). PTSD = posttraumatic stress disorder; L = left; R = right. www.nature.com/scientificreports/ Table 3. PTSD symptom clusters as predictors of amygdala and hippocampus grey matter volume. Within the PTSD group, greater hyperarousal symptoms significantly predicted less left amygdala volume, even after controlling for anxiety and depression (denoted *); while greater avoidance symptoms significantly predicted greater right amygdala volume, even after controlling for anxiety and depression (denoted *). PTSD symptom cluster scores within the PTSD group did not significantly (all ps > .0125) predict left or right hippocampal volume. TIV = total intracranial volume.
Step 1 Step 2 Step 3 Step 4 www.nature.com/scientificreports/ In conclusion, the current study found no differences in CT or GMV in the amygdala or hippocampus of adult women with PTSD stemming from IPV-exposure, compared to trauma-free women without a diagnosis of PTSD. Although there were no group differences in the current study, re-experiencing, active avoidance, and hyperarousal symptom severity within the PTSD group were significantly associated with differential (increased and decreased depending on the region) CT in several regions and GMV in the amygdala. These findings add to the growing literature suggesting that: (1) the specific type of trauma exposure (e.g., interpersonal violence vs combat) may play an important role in determining whether a diagnosis of PTSD predicts structural brain differences compared to trauma-free controls, (2) alterations in CT and GMV among women with PTSD stemming from IPV-exposure may be driven by symptom severity within specific symptom clusters, as opposed to overall PTSD symptom severity, and (3) symptom severity within PTSD symptom clusters may have a differential effect on brain structure among women with IPV-related PTSD, with greater avoidance symptom severity predicting increased CT and GMV, and greater re-experiencing and hyperarousal symptom severity predicting decreased CT and GMV.