Network-level structural alterations distinguish persistent from remitted post-traumatic stress disorder: a morphometric covariance approach Authors: Suji Lee, Hyeonji Lee, Yoonji Joo, Chaewon Suh, Youngeun Shim, Yunjung Jin, Harin Song, Sujung Yoon, In Kyoon Lyoo Published: 2026 ABSTRACT ABSTRACTBackground: Post-traumatic stress disorder (PTSD) exhibits heterogeneous clinical trajectories: while many affected individuals achieve remission, a substantial proportion develop a chronic, persistent form. The neuroanatomical mechanisms distinguishing these divergent outcomes remain poorly defined. This study aimed to elucidate the structural network organisation of persistent PTSD using structural covariance network analysis.Methods: Participants included trauma-unexposed healthy controls (HC; n = 112), individuals with remitted PTSD (n = 32), and those with persistent PTSD (n = 78). High-resolution T1-weighted magnetic resonance imaging data were analyzed using independent component analysis to extract spatially independent morphometric networks. Gray matter volumes within these networks were subsequently compared across groups.Results: Linear trend analysis revealed a stepwise reduction in gray matter volume within the frontal-cognitive control and visual networks across groups (HC > remitted PTSD > persistent PTSD). Further examination of inter-network connectivity demonstrated heightened connectivity between the frontal-cognitive control and temporo-insular-limbic networks in the persistent PTSD group, compared with both the remitted PTSD and HC groups. Notably, the remitted PTSD group exhibited inter-network connectivity levels comparable to those of HC. This hyperconnectivity in persistent PTSD reflects compensatory recruitment of cognitive control mechanisms to regulate interoceptive and emotional processing systems, thereby sustaining symptom modulation.Conclusion: Persistent PTSD is characterised by progressive gray matter volume reductions and compensatory hyperconnectivity between cognitive control and limbic-interoceptive networks, indicative of sustained engagement of regulatory mechanisms to modulate enduring symptoms. These findings provide novel neurobiological insights into the mechanisms of PTSD chronicity and highlight potential network-level targets (particularly prefrontal-limbic interactions) for future investigation of personalised interventions in treatment-resistant cases.KEYWORDS: Persistent post-traumatic stress disorder, structural covariance, morphometric networks, frontal-cognitive control network, compensatory hyperconnectivity Introduction Post-traumatic stress disorder (PTSD) demonstrates highly heterogeneous clinical trajectories following trauma exposure. While a substantial proportion of individuals develop chronic, treatment-refractory symptoms, others exhibit resilience or recovery, either maintaining subclinical manifestations or transitioning from an initial PTSD diagnosis to full remission (Bryant et al., 2023; Karchoud et al., 2025; Neria, 2021). This divergence raises fundamental questions regarding the neurobiological mechanisms that differentiate persistent symptomatology from recovery. Elucidating the neuroanatomical substrates underlying these contrasting trajectories is of critical theoretical and clinical importance. Identifying structural brain alterations that distinguish persistent from remitted PTSD could illuminate the pathophysiological mechanisms associated with chronic symptoms, advance understanding of trauma-related neuroplasticity, and guide the development of targeted therapeutic interventions (Logue et al., 2018; Zilcha-Mano et al., 2022). Such insights may enable clinicians to identify individuals at risk for persistent symptoms, optimise treatment strategies based on neurobiological profiles, and design interventions that specifically target maladaptive network-level alterations underlying treatment resistance. Neurobiological models of PTSD consistently implicate dysfunction within a core network of brain regions – including the prefrontal cortex, amygdala, and hippocampus – in the disorder's pathophysiology (Logue et al., 2018; Pitman et al., 2012; Zilcha-Mano et al., 2022). However, most previous studies have examined PTSD cohorts as a homogeneous group, without directly contrasting persistent and remitted trajectories. As a result, although the general neural correlates of PTSD are increasingly recognised, the specific neuroanatomical alterations associated with the symptom chronicity remain poorly defined. Evidence from recovery-focused studies suggests that clinical remission may be accompanied by structural profiles comparable to healthy controls of key regions, particularly the prefrontal cortex, which plays a central role in the top-down regulation of emotional responses (Boukezzi et al., 2017; Hinojosa et al., 2024; Lyoo et al., 2011). This supports the hypothesis that persistent dysfunction within prefrontal regulatory systems may constitute a critical neurobiological correlate of PTSD chronicity. Based on this, we formed an a priori hypothesis that PTSD chronicity imposes a graded, ‘dose-dependent’ effect on brain structure. We posited that remitted individuals might exhibit residual neuroanatomical ‘scarring’ intermediate between healthy controls and persistent cases, reflecting the cumulative burden of prior trauma and stress exposure (Franz et al., 2020; Heyn & Herringa, 2019; Roberts et al., 2022). However, we also considered alternative models raised in the literature, including structural restoration upon remission or volumetric reductions confined solely to persistent PTSD. Yet, the prefrontal cortex does not function in isolation; its regulatory influence is executed through integration within large-scale brain networks (Negrón-Oyarzo et al., 2024; van Es et al., 2025). Consequently, traditional neuroimaging approaches focusing on discrete brain regions may be insufficient to capture the distributed, network-level disruptions underlying this complex clinical phenomenon. To overcome these limitations, a network-based perspective is essential. Psychiatric disorders are increasingly understood to arise from alterations in the coordinated organisation of large-scale brain systems rather than from isolated regional deficits (Bassett & Sporns, 2017; Ishida et al., 2023). Structural covariance analysis provides a powerful framework for this investigation, as it evaluates correlations in gray matter morphometry to infer patterns of anatomical connectivity and shared pathological influences (Alexander-Bloch et al., 2013). Within this context, independent component analysis (ICA) offers a robust, data-driven method to identify spatially independent morphometric networks based on covariance patterns across individuals (Mongay-Ochoa et al., 2025; Xu et al., 2009). This methodology enables the delineation of functionally meaningful brain networks and facilitates the direct investigation of network-level differences between persistent and remitted PTSD that may not be evident with traditional regional analyses. The present study sought to elucidate the structural brain network organisation underlying persistent and remitted PTSD through morphometric network analysis. Specifically, ICA was applied to high-resolution neuroimaging data to extract data-driven morphometric networks, enabling assessment of both regional gray matter volumes and inter-network structural covariance across three groups: trauma-unexposed healthy controls, individuals with remitted PTSD, and those with persistent PTSD. Given the prefrontal cortex's pivotal role in top-down emotional regulation (Hinojosa et al., 2024; Pitman et al., 2012) and its involvement in recovery processes (Boukezzi et al., 2017; Lyoo et al., 2011; Shaw et al., 2023), we hypothesised that perturbations within prefrontal-related networks – particularly those mediating cognitive control and emotional regulation – would distinguish persistent PTSD from remitted cases. Materials and methods Participants A total of 224 women were enrolled in this study, comprising 112 women individuals who had been diagnosed with PTSD and 112 demographically-matched healthy controls with no history of trauma exposure. This cohort was drawn from a previously published sample (Lee et al., 2023). To ensure homogeneity of trauma type – thereby minimising potential confounding from heterogeneous etiological factors and enhancing the interpretability of neurobiological findings – the trauma-exposed group consisted exclusively of female victims of intimate partner violence (IPV), including physical, emotional, and/or sexual assaults perpetrated by a spouse or intimate partner (Krantz & Garcia-Moreno, 2005). These participants were recruited through advertisements placed in trauma counselling centres and police stations within the metropolitan areas of Seoul, South Korea. PTSD symptom severity was assessed using the Clinician-Administered PTSD Scale (CAPS) for the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) (Blake et al., 1995). The trauma-unexposed healthy control group was recruited through local advertisements, complemented by snowball sampling. All participants underwent comprehensive psychiatric evaluations conducted by board-certified psychiatrists. Exclusion criteria included any major psychiatric disorders predating the index trauma, clinically significant medical or neurological conditions, or contraindications for magnetic resonance imaging (MRI). Detailed demographic and clinical characteristics are provided in Table 1. Table 1.Demographic and clinical sample characteristics.CharacteristicsTrauma-exposed individuals(n = 112)Control(n = 112)p-valueaPersistent PTSD(n = 78)Remitted PTSD(n = 32)Demographic characteristics Age, mean (SD), years41.2 (12.2)39.5 (13.1)41.1 (12.7).789 Female, No. (%)78 (100)32 (100)112 (100)– High school completion or higher, No. (%)68 (87.2)29 (90.6)105 (93.8).256 Right handedness, No. (%)74 (94.9)27 (84.4)104 (92.9).194Trauma exposure-related characteristics Time since trauma, mean (SD), months47.7 (97.0)66.9 (105.8)NA.361 Age at index trauma, mean (SD), years37.3 (12.8)34.0 (14.5)NA.248 Trauma Recurrence, n (%) .583 Single episode25 (32.1)12 (37.5)NA Repeated exposure53 (68.0)20 (62.5)NA Trauma type, No (%)b NA Physical violence66 (84.6)28 (87.5)NA.776 Rape15 (19.2)4 (12.5)NA.580 Sexual violence6 (7.69)5 (15.6)NA.292 Psychological violence/threats75 (96.2)30 (93.8)NA.627 CAPS total scores, mean (SD)36.6 (8.8)13.3 (5.6)NA<.001 Use of psychotropic medications for at least 2 weeks, No (%)36 (46.2)9 (28.1)NA.081 Comorbid psychiatric conditions, No (%)c Mood Disorders23 (29.5)5 (15.6)NA.154 Anxiety Disorders4 (5.13)1 (3.13)NA1.00 Substance Use Disorders6 (7.69)0 (0)NA.178 Others2 (2.56)0 (0)NA1.00Note. Data are presented as mean (SD) or number (%).aDemographic and clinical characteristics across groups were examined using one-way ANOVA, independent-samples t-tests, or Fisher’s exact tests, as appropriate.bTrauma types are not mutually exclusive; participants may have experienced multiple types.cComorbid psychiatric conditions include Mood Disorders (major depressive disorder, bipolar disorder, dysthymic disorder), Anxiety Disorders (panic disorder, specific phobia), Substance Use Disorders (alcohol dependence, alcohol abuse), and Others (schizophrenia, eating disorders). Categories are not mutually exclusive, as some participants presented with multiple comorbid conditions.Abbreviations: ANOVA, analysis of variance; CAPS, Clinician Administered PTSD Scale for Diagnostic and Statistical Manual of Mental Disorders, fifth edition; Control, trauma-unexposed individuals; NA, not applicable; No, number; PTSD, post-traumatic stress disorder; SD, standard deviation. Demographic and clinical sample characteristics. Note. Data are presented as mean (SD) or number (%). Demographic and clinical characteristics across groups were examined using one-way ANOVA, independent-samples t-tests, or Fisher’s exact tests, as appropriate. Trauma types are not mutually exclusive; participants may have experienced multiple types. Comorbid psychiatric conditions include Mood Disorders (major depressive disorder, bipolar disorder, dysthymic disorder), Anxiety Disorders (panic disorder, specific phobia), Substance Use Disorders (alcohol dependence, alcohol abuse), and Others (schizophrenia, eating disorders). Categories are not mutually exclusive, as some participants presented with multiple comorbid conditions. Abbreviations: ANOVA, analysis of variance; CAPS, Clinician Administered PTSD Scale for Diagnostic and Statistical Manual of Mental Disorders, fifth edition; Control, trauma-unexposed individuals; NA, not applicable; No, number; PTSD, post-traumatic stress disorder; SD, standard deviation. The trauma-exposed participants were further stratified based on diagnostic status at the time of assessment into a persistent PTSD subgroup (n = 78) and a remitted PTSD subgroup (n = 34). Two participants from the remitted subgroup were excluded due to suboptimal image quality, resulting in a final sample of 32 individuals in this subgroup. In the persistent PTSD group (n = 78), 36 individuals (46.2%) reported psychotropic medication use for a minimum of two weeks, including antidepressants (n = 26, 33.3%), mood stabilisers (n = 2, 2.56%), and anxiolytics/sedatives (n = 35, 44.9%). In the remitted PTSD group (n = 32), 9 participants (28.1%) reported psychotropic medication use, comprising antidepressants (n = 8, 25.0%), mood stabilisers (n = 2, 6.25%), and anxiolytics/sedatives (n = 9, 28.1%). Both groups displayed comparable trauma exposure characteristics, with no significant differences in repeated exposure history, time since trauma, or age at trauma exposure (Table 1). Comorbid psychiatric conditions were present in both cohorts but did not differ significantly between groups: mood disorders (persistent: 29.5%; remitted: 15.6%), anxiety disorders (5.13% vs. 3.13%), and substance use disorders (7.69% vs. 0%) (Table 1). The study protocol was approved by the Institutional Review Board of Ewha Womans University, and all participants provided written informed consent. Image acquisition and preprocessing Brain MRI data were acquired at the Ewha Brain Institute, Ewha Womans University, Seoul, South Korea, using a 3.0 Tesla Philips scanner (Philips Medical Systems, Best, The Netherlands) equipped with a 32-channel head coil. High-resolution T1-weighted anatomical images were obtained with the following parameters: repetition time (TR) = 7.4 ms; echo time (TE) = 3.4 ms; field of view (FOV) = 224 × 224 mm2; slice thickness = 1 mm; flip angle = 8°; and 180 sagittal slices. In addition, fluid-attenuated inversion recovery (FLAIR) images were acquired with the following parameters: TR = 4,800 ms; TE = 278 ms; inversion time = 1,650 ms; FOV = 240 × 240 mm2; matrix = 216 × 216; voxel size = 1.11 × 1.11 × 1.12 mm3. Both T1-weighted and FLAIR images were reviewed by an experienced radiologist to exclude overt brain abnormalities. As mentioned above, due to suboptimal image quality, T1-weighted MRI scans from two participants in the remitted PTSD group were excluded from further analyses. Gray matter volumes were estimated using voxel-based morphometry (VBM) implemented in the FSL-VBM pipeline (version 6.0.4; FMRIB Software Library), which follows an optimised VBM protocol (Douaud et al., 2007; Good et al., 2001; Smith et al., 2004). Preprocessing involved several steps. First, non-brain tissue was removed from the T1-weighted images using FSL's automated Brain Extraction Tool. Next, tissue segmentation was performed to classify the brain into gray matter, white matter, and cerebrospinal fluid. Partial volume estimation was applied to account for voxels containing multiple tissue types, yielding probabilistic gray matter maps rather than binary segmentations. For spatial normalisation, native gray matter images were non-linearly registered to the Montreal Neurological Institute (MNI)152 template. These aligned images were averaged to create a study-specific gray matter template, ensuring unbiased registration across participants. All native gray matter images were then non-linearly re-registered to this study-specific template. To preserve total gray matter volume, the registered maps were modulated by multiplying voxel intensities by the Jacobian determinant of the warp field. Finally, the modulated gray matter images were smoothed using an isotropic Gaussian kernel with a sigma of 3 mm to increase signal-to-noise ratio and minimise residual misregistration effects (Coppen et al., 2016; Hafkemeijer et al., 2014; Yang et al., 2021). To control for inter-individual differences in head size, total intracranial volume (ICV) was extracted using FreeSurfer (Buckner et al., 2004). Independent component analysis and structural covariance network analysis Preprocessed gray matter volume maps from all participants were concatenated and subjected to group-level ICA using FSL's Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC; https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC). This approach identified spatially independent morphometric networks based on structural covariance patterns across participants, reflecting coordinated gray matter volume variations that may arise from shared developmental, genetic, or pathological influences (Alexander-Bloch et al., 2013). The analysis was configured to extract 30 spatially independent components, a dimensionality selected based on validation studies demonstrating optimal decomposition of structural covariance patterns in comparable sample sizes (Joo et al., 2023; Pichet Binette et al., 2020; Yang et al., 2021; Zeighami et al., 2015). Each independent component (IC) was transformed into standardised z-statistics and thresholded at z > 3.5 to retain only voxels contributing significantly to the component's spatial pattern. The resulting thresholded ICs were defined as ‘morphometric networks,’ representing anatomically coherent regions exhibiting coordinated structural variation across participants. To facilitate interpretation and reduce dimensionality, morphometric networks were further organised into ‘network modules’ based on their spatial correspondence with established resting-state functional networks, using validated parcellation schemes (Luo et al., 2020; Rodrigue et al., 2020). Individual-specific gray matter volumes were then extracted from each network module by averaging voxel values within module boundaries, providing subject-level measures for subsequent group comparisons. Inter-network connectivity was quantified by calculating partial correlation coefficients between gray matter volumes of each network module pair within each group, controlling for age (Joo et al., 2023). These coefficients were Fisher r-to-z transformed to ensure normal distribution prior to statistical testing (Cohen et al., 2003). Statistical analysis Demographic and clinical characteristics were compared across groups using one-way analysis of variance or independent t-tests for continuous variables and chi-square tests for categorical variables. Group differences in mean gray matter volumes for each network module were evaluated using linear trend analysis, with group membership (healthy controls, remitted PTSD, and persistent PTSD) treated as an ordinal variable to test for progressive patterns. Age and total ICV were included as covariates to mitigate confounding and improve the precision of estimates. To complement the linear trend and explore alternative patterns (e.g. reductions specific to persistent PTSD or restoration in remission), we conducted pairwise group comparisons using dummy-coded regression analyses, with gray matter volume as the dependent variable and group as a categorical predictor, controlling for age and ICV. Standardised beta coefficients were estimated to quantify effect sizes. Contrasts tested differences between healthy controls vs. remitted PTSD, healthy controls vs. persistent PTSD, and remitted PTSD vs. persistent PTSD. Partial correlation analyses were performed to examine associations between gray matter volumes of network modules and PTSD symptom severity, adjusting for age and total ICV. Group differences in inter-network connectivity were assessed by comparing Fisher z-transformed partial correlation coefficients between groups using independent two-sample Z-tests (Cohen et al., 2003; Joo et al., 2023; Khundrakpam et al., 2017; Lee et al., 2024; Pichet Binette et al., 2020). Pairwise comparisons were conducted for all group pairings (healthy controls vs. remitted PTSD, healthy controls vs. persistent PTSD, remitted PTSD vs. persistent PTSD) across each network module pair. To correct for multiple testing, p-values were adjusted using the false discovery rate (FDR) procedure (Genovese et al., 2002). All statistical procedures were conducted using R software, version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria) and Stata (version 16.0; StataCorp, College Station, TX, USA). Results Morphometric networks and network modules A data-driven ICA was applied to gray matter volume maps from all participants, yielding 30 statistically independent morphometric networks. Following visual inspection and quality assessment, one component was identified as an artifact and excluded, resulting in 29 valid morphometric networks. These networks were subsequently aggregated into five primary network modules, with their composite spatial maps presented in Figure 1. Spatial maps of five network modules and their constituent independent components (morphometric networks). Spatial maps display 29 morphometric networks derived from group-level independent component analysis (ICA) applied to high-resolution T1-weighted structural MRI data from 222 participants. These 29 spatially independent components were organised into five functionally coherent network modules based on their anatomical correspondence with established brain systems: (1) frontal-cognitive control network, primarily distributed in the prefrontal cortex and anterior cingulate regions; (2) default mode network, spanning the medial prefrontal cortex, posterior cingulate cortex, and angular gyrus; (3) temporo-insular-limbic network, comprising the temporal cortex, insular regions, basal ganglia, hippocampus, and amygdala; (4) visual network, localised to occipital and visual cortical areas; and (5) sensorimotor network, including the primary motor and somatosensory cortices. Network modules were overlaid onto cortical surfaces (left panels), and cross-sectional slices illustrate morphometric network components in Montreal Neurological Institute (MNI) coordinate space (right panels). The colour bar represents z-score values from the ICA decomposition. Abbreviations: L, left hemisphere; MRI, magnetic resonance imaging; R, right hemisphere. The identified modules included the frontal-cognitive control network (ICs 5, 7, 13, 23), the default mode network (ICs 8, 14, 16, 17, 26, 27, 28), the temporo-insular-limbic network (ICs 1, 2, 10, 29, 30), the visual network (ICs 4, 6, 9, 15, 18, 20, 21), and the sensorimotor network (ICs 3, 11, 12, 19, 22, 25). Stepwise reduction in gray matter volumes across groups Stepwise reductions in gray matter volume were examined across the three groups (healthy controls > remitted PTSD > persistent PTSD) for each network module. Linear trend analysis, controlling for age and total ICV, revealed significant progressive reductions in two network modules. The frontal-cognitive control network showed a significant negative linear trend (β = −0.01, p = .017, FDR-corrected p < .05), indicating systematic volumetric decreases from healthy controls through remitted PTSD to persistent PTSD. Similarly, the visual network demonstrated a comparable pattern of progressive reduction (β = −0.01, p = .019, FDR-corrected p < .05), as illustrated in Figure 2. Mean gray matter volumes across network modules and diagnostic groups. Bar graphs illustrate mean gray matter volumes in the (a) frontal-cognitive control network, (b) default mode network, (c) temporo-insular-limbic network, (d) visual network, and (e) sensorimotor network, compared across healthy controls, remitted PTSD, and persistent PTSD groups. Asterisks (*) indicate significant linear trends (p < .05, false discovery rate-corrected), reflecting progressive volume reductions (healthy controls > remitted PTSD > persistent PTSD). Error bars denote the standard error of the mean for each group. Abbreviations: GM, gray matter; L, left hemisphere; PTSD, post-traumatic stress disorder; R, right hemisphere. In contrast, no statistically significant linear trends were observed in the other network modules: default mode network (β < −0.01, p = .061), temporo-insular-limbic network (β < −0.01, p = .200), and sensorimotor network (β < −0.01, p = .095). Pairwise comparisons, controlling for age and ICV, confirmed significant gray matter reductions in persistent PTSD relative to healthy controls for the frontal-cognitive control network (β = −0.17, p = .017) and visual network (β = −0.16, p = .020). No significant differences were found between remitted PTSD and healthy controls (frontal-cognitive: β = −0.06, p = .404; visual: β = −0.07, p = .292) or between remitted and persistent PTSD (frontal-cognitive: β = −0.09, p = .383; visual: β = −0.06, p = .533). For the default mode network, persistent PTSD showed nominally lower volumes compared with HC, though this contrast did not reach conventional statistical significance (β = −0.13, p = .055). There were no other significant pairwise differences (remitted PTSD vs. healthy controls, β < 0.01, p = .965; remitted PTSD vs. persistent PTSD, β = −0.13, p = .166). Pairwise comparisons were non-significant for the sensorimotor (persistent PTSD vs. healthy controls, β = −0.12, p = .092; remitted PTSD vs. healthy controls, β = −0.02, p = .757; remitted PTSD vs. persistent PTSD, β = −0.09, p = .376) and temporo-insular-limbic networks (persistent PTSD vs. healthy controls, β = −0.09, p = .182; remitted PTSD vs. healthy controls, β = 0.02, p = .758; remitted PTSD vs. persistent PTSD, β = −0.12, p = .220). Partial correlation analyses, adjusted for age and total ICV, were conducted separately within the remitted and persistent PTSD groups. These analyses revealed no significant associations between PTSD symptom severity – as measured by CAPS total score – and gray matter volumes across any of the network modules: frontal-cognitive control network (remitted: r = 0.08, p = .67; persistent: r = −0.08, p = .48), default mode network (remitted: r = 0.06, p = .74; persistent: r = −0.11, p = .37), temporo-insular-limbic network (remitted: r = 0.22, p = .25; persistent: r = −0.05, p = .65), visual network (remitted: r = 0.13, p = .49; persistent: r = 0.03, p = .78), and sensorimotor network (remitted: r = 0.14, p = .47; persistent: r = −0.01, p = .95). Altered inter-network connectivity of the frontal-cognitive control network in persistent PTSD Given our a priori hypothesis that alterations in prefrontal systems contribute critically to PTSD chronicity (Boukezzi et al., 2017; Lee et al., 2023; Lyoo et al., 2011), we examined group differences in structural covariance, focusing on the connectivity of the frontal-cognitive control network with the other four network modules. Inter-network connectivity between the frontal-cognitive control network and the remaining four network modules was compared across groups. Specifically, connectivity between the frontal-cognitive control network and the temporo-insular-limbic network was significantly greater in the persistent PTSD group compared with both healthy controls (z = −2.85, FDR-corrected p < .05) and the remitted PTSD group (z = 3.00, FDR-corrected p < .05), as shown in Figure 3. No significant differences in this connectivity were observed between the remitted PTSD group and healthy controls (z = 1.10, FDR-corrected p > .05). Inter-network connectivity of the frontal-cognitive control network module with other brain network modules across groups. (a) Schematic configuration of brain network modules centred on the frontal-cognitive control network, illustrating constituent brain regions and baseline inter-network connectivity profiles between the frontal-cognitive control network and other functional modules in healthy controls. (b) Radar plots depict inter-network connectivity profiles between the frontal-cognitive control network and the other four network modules across healthy controls, remitted PTSD, and persistent PTSD groups. (c) Bar graph shows group differences in connectivity strength between the frontal-cognitive control and temporo-insular-limbic network modules. Persistent PTSD exhibited significantly stronger connectivity compared with both healthy controls and remitted PTSD, whereas no differences were observed between remitted PTSD and controls. Asterisks (*) indicate FDR-corrected P < 0.05. Error bars denote the standard error of the mean. Abbreviations: DMN, default mode network; FrontCogcontrol, frontal-cognitive control network; PTSD, post-traumatic stress disorder; SMN, sensorimotor network; TempInsLim, temporo-insular-limbic network; VIN, visual network. Connectivity between the frontal-cognitive control network and the remaining three modules (default mode, visual, and sensorimotor networks) did not differ significantly across persistent PTSD, remitted PTSD, and healthy control groups. Discussion This study employed a comprehensive morphometric network analysis to investigate structural brain organisation in persistent versus remitted PTSD. Two key neurobiological signatures emerged that differentiate persistent PTSD from both remitted PTSD and healthy controls: (1) progressive gray matter volume reductions in the frontal-cognitive control and visual networks, following a stepwise pattern across diagnostic categories, and (2) selective hyperconnectivity between the frontal-cognitive control and temporo-insular-limbic networks specific to persistent PTSD. Notably, the remitted PTSD group exhibited connectivity patterns comparable to healthy controls, suggesting normalisation of inter-network communication with symptom resolution. The stepwise gray matter volume reductions observed in the frontal-cognitive control network are consistent with theories implicating prefrontal dysfunction in PTSD pathophysiology (Pitman et al., 2012). Importantly, these findings extend prior work by demonstrating that structural alterations parallel clinical trajectories, with the most pronounced reductions in persistent PTSD. This graded pattern suggests that prefrontal structural integrity may represent a neurobiological marker of symptom persistence, rather than a general correlate of trauma exposure. Pairwise comparisons further elucidated that these reductions are primarily driven by differences between persistent PTSD and healthy controls, with remitted PTSD displaying intermediate volumes that do not significantly diverge from either group. This pattern supports interpretations of either a linear trend reflecting cumulative ‘scarring’ from trauma exposure (Franz et al., 2020; Heyn & Herringa, 2019; Roberts et al., 2022) or alternative trajectories, such as structural restoration in remission (Hinojosa et al., 2024; Stoklosa et al., 2024) or volumetric reductions confined to persistent cases (Hinojosa et al., 2024; Tanev et al., 2025). For the default mode network, the borderline pairwise reduction in persistent PTSD versus controls suggests subtle involvement that may become evident in larger samples, potentially indicating broader network vulnerability in chronicity. The mechanisms underlying these reductions likely reflect multiple pathophysiological processes. One prominent hypothesis is that chronic exposure to stress hormones and excitotoxicity following trauma drives progressive prefrontal atrophy, thereby impairing top-down regulation of fear and stress responses (Akiki et al., 2017). The intermediate volumes observed in remitted PTSD indicate that residual structural ‘scarring’ or plasticity-related modifications may persist beyond symptom resolution. This finding is consistent with the notion that selective neuroanatomical alterations endure post-remission, potentially serving as biomarkers of prior trauma exposure or indicators of relapse vulnerability (Sun et al., 2018). However, the cross-sectional nature of our design precludes determination of whether these patterns reflect pre-existing vulnerability factors that predispose certain individuals to chronic PTSD trajectories (Ben-Zion et al., 2023) or consequences of PTSD chronicity – a question that requires longitudinal investigation. It is noteworthy that no significant associations were observed between current PTSD symptom severity – as measured by CAPS total score – and gray matter volumes across any of the network modules within either patient group. The absence of a dose–response relationship is consistent with a growing body of evidence in the PTSD neuroimaging literature, where structural alterations are reliably observed between patients and healthy controls but show weak or inconsistent correlations with concurrent symptom severity within patient samples (Akiki et al., 2017; Crombie et al., 2021; Huggins et al., 2024; Siehl et al., 2023). This pattern has frequently been interpreted as suggesting that such volumetric reductions may represent relatively stable, trait-like abnormalities or the cumulative burden of chronic PTSD and repeated stress exposure, rather than dynamic state markers that closely track fluctuations in current symptom severity. While our cross-sectional data cannot confirm causality or temporal dynamics, the lack of association with current CAPS scores is compatible with this view and warrants further investigation in longitudinal designs to disentangle trait versus state contributions to structural network changes. Progressive volume reductions were also identified in the visual network, encompassing occipital and temporal regions integral to sensory processing. This finding corroborates emerging evidence implicating the visual cortex in PTSD, potentially through heightened perceptual sensitivity or intrusive visual memories (Chao et al., 2012; Kimble et al., 2010; Mueller-Pfeiffer et al., 2013). As with the frontal-cognitive network, pairwise analyses revealed significant differences primarily between persistent PTSD and controls, reinforcing a chronic-specific mechanism while allowing for interpretations of restoration in remission. The involvement of the visual network may reflect trauma-specific mechanisms, such as visual threats associated with IPV, which could drive adaptive or maladaptive plasticity in sensory processing pathways (Clauss & Clements, 2021; Tomoda et al., 2012). By contrast, the absence of linear trends in the default mode, temporo-insular-limbic, and sensorimotor networks indicates that PTSD chronicity is not uniformly distributed across brain systems, but instead selectively targets networks central to cognitive-emotional regulation and perceptual vigilance. The selective hyperconnectivity between the frontal-cognitive control and temporo-insular-limbic networks represents a novel finding that offers mechanistic insights into PTSD chronicity. This connectivity pattern was observed exclusively in the persistent PTSD group, distinguishing it from both remitted PTSD and healthy controls. The temporo-insular-limbic network, encompassing temporal, insular, limbic, and basal ganglia structures, includes regions essential for fear conditioning, interoceptive awareness, emotional salience attribution, and autonomic regulation – domains consistently dysregulated in PTSD (Paulus & Stein, 2010; Pitman et al., 2012; Yoon et al., 2017). These results are consistent with contemporary neurobiological models suggesting that the prefrontal cortex exerts regulatory influence not in isolation but through integration within large-scale brain networks (Menon & D'Esposito, 2022). The observed hyperconnectivity may reflect compensatory neural recruitment, wherein cognitive control systems are engaged at heightened levels to regulate dysregulated interoceptive and emotional signals – a pattern supported by functional connectivity studies in chronic PTSD (Fonzo et al., 2022; Keller et al., 2022; Nicholson et al., 2020). However, alternative interpretations are equally plausible, such as maladaptive over-engagement of regulatory systems leading to inefficiency or resource depletion (Akiki et al., 2017; Haris et al., 2023), failed top-down control signalling persistent regulatory breakdown (Sripada et al., 2012), or shared pathological influences (e.g. common neurodegenerative or inflammatory processes) driving correlated morphometric changes across both networks (Alexander-Bloch et al., 2013; Paulus & Stein, 2010). Functional neuroimaging studies are warranted to disentangle whether this pattern corresponds to adaptive compensation or maladaptive dysregulation. Crucially, these inter-network perturbations – undetectable through conventional regional analyses – underscore the value of covariance-based methodologies for identifying distributed pathological processes (Alexander-Bloch et al., 2013; Bassett & Sporns, 2017). Although the salience network is a well-characterised core system in PTSD functional neuroimaging literature (Roeckner et al., 2021; Zhang et al., 2022), its components were distributed across the temporo-insular-limbic (insula, amygdala) and frontal-cognitive control (dACC) modules in our ICA-derived SCN, reflecting data-driven covariance patterns rather than a priori functional labels. This dissociation is consistent with prior evidence that structural covariance networks do not necessarily replicate the boundaries of canonical functional networks (Lizarraga et al., 2024). Nevertheless, the hyperconnectivity between the frontal-cognitive control and temporo-insular-limbic networks observed in persistent PTSD may reflect altered communication within the salience system at a structural level. This pattern aligns with prevailing models of salience network dysfunction in PTSD, which posit aberrant integration between cognitive control and interoceptive-emotional processing systems (Dimitrova et al., 2025; Haris et al., 2024; Rakesh et al., 2023). In a supplementary analysis, we examined gray matter volume alterations in individual ICs most closely corresponding to salience network regions (e.g. insula-dominant IC 29, dACC-dominant IC 7, amygdala-dominant IC 30). No significant group differences were observed, suggesting that salience-related pathophysiology in persistent PTSD may manifest primarily through altered inter-network communication rather than isolated regional volume loss. This interpretation is consistent with our main finding of hyperconnectivity between frontal-cognitive control and temporo-insular-limbic networks. Therefore, this null finding highlights the complementary value of functional connectivity analyses, which could better capture dynamic salience dysregulation and validate our structural results in future studies. The connectivity patterns observed in the remitted PTSD group – comparable to those of healthy controls – support a recovery model characterised by the presence of balanced inter-network communication. Symptom resolution may be associated with balanced network integration through neuroplastic processes, an interpretation consistent with functional neuroimaging studies demonstrating reduced pathological prefrontal-limbic coupling following effective treatment (Fonzo et al., 2017) and evidence that remitted individuals show structural and functional patterns within prefrontal circuits comparable to healthy controls (Hinojosa et al., 2024; Yang et al., 2021). Nevertheless, given the cross-sectional nature of our design, we cannot rule out the possibility that intact connectivity represents a pre-existing protective factor that facilitated symptom remission rather than a consequence of recovery. Individuals with inherently resilient network architectures may possess greater capacity for emotional regulation, predisposing them to better clinical outcomes following trauma exposure (Roeckner et al., 2021). The observed pattern of progressive gray matter volume reductions in the frontal-cognitive control network combined with selective hyperconnectivity to the temporo-insular-limbic system in persistent PTSD suggests that chronicity may involve both structural compromise and compensatory over-recruitment of cognitive regulatory resources to manage persistent limbic-interoceptive dysregulation. While our cross-sectional design precludes direct causal inferences, these network-level signatures point toward plausible therapeutic directions for treatment-resistant cases, where conventional pharmacotherapy and trauma-focused psychotherapy may yield suboptimal outcomes. Specifically, interventions that enhance prefrontal cognitive control over limbic reactivity or normalise fronto-limbic connectivity represent promising avenues. Repetitive transcranial magnetic stimulation (rTMS) targeting the dorsolateral prefrontal cortex, a core node of the frontal-cognitive control network, has shown promise in reducing PTSD symptoms by augmenting top-down regulation and decreasing amygdala hyperreactivity (Brown et al., 2025). Real-time fMRI neurofeedback training, which enables patients to volitionally modulate prefrontal-limbic connectivity during symptom provocation, has also demonstrated feasibility and symptom reduction in PTSD by enhancing regulatory control over limbic hyperactivation (Voigt et al., 2024; Zhao et al., 2023). Cognitive training paradigms focused on executive functions (e.g. working memory or response inhibition tasks) could similarly target the frontal-cognitive control network to bolster endogenous regulation of interoceptive and emotional signals (Bomyea et al., 2025). Finally, emerging closed-loop neuromodulation approaches, which use real-time biomarkers (e.g. amygdala theta power) to trigger stimulation, offer potential for precision targeting of the identified hyperconnectivity pattern in treatment-resistant individuals (Gill et al., 2023). Although these strategies remain investigational, our findings provide a neurobiological rationale for prioritising interventions that restore balanced prefrontal-limbic integration in chronic PTSD. The present findings carry several important clinical implications. First, the identification of distinct neurobiological signatures differentiating persistent from remitted PTSD may inform prognostic evaluation and treatment planning. Individuals exhibiting hyperconnectivity between cognitive control and limbic networks may particularly benefit from interventions aimed at enhancing prefrontal regulatory efficiency, such as cognitive training or neuromodulatory strategies. Second, the observation that remitted PTSD is associated with normalised connectivity patterns indicates that structural network organisation may serve as a biomarker of treatment response. Longitudinal studies are warranted to determine whether changes in inter-network connectivity predict treatment outcomes or provide early markers of recovery. Several methodological considerations should be noted. The cross-sectional design precludes definitive causal inferences regarding the temporal relationship between structural alterations and symptom trajectories. Prospective longitudinal studies following individuals from acute trauma exposure into chronic phases will be critical for elucidating the developmental course of these network alterations. In addition, the study sample consisted exclusively of female IPV survivors. While this enhances trauma type homogeneity, it limits generalizability across trauma types and genders. Future research should examine whether similar network patterns emerge across diverse trauma populations. Although the traumatic source in our sample was IPV, a characteristically chronic and recurrent form of trauma, detailed information on the precise duration and frequency of IPV exposure was not available. While proxy measures such as trauma repeatedness and trauma types did not significantly differ between groups, these may not fully capture IPV severity and chronicity. Additionally, ‘time since trauma’ refers to the interval since the index trauma precipitating the current PTSD episode, and does not account for potential subthreshold IPV exposure prior to this event. Furthermore, comorbid psychiatric conditions and psychotropic medication use could represent important confounding factors. Both groups exhibited statistically balanced profiles regarding the prevalence of comorbid mood, anxiety, and substance use disorders. Similarly, while a history of psychotropic medication use was numerically more common in the persistent group (46.2%) than in the remitted group (28.1%), this difference was not statistically significant. Collectively, while we cannot fully exclude the possibility that cumulative trauma load or specific comorbidities independently modulate brain structure, the lack of significant group differences in these key variables – though not definitively ruling out their influence – provides preliminary support for the interpretation that the observed network-level alterations are primarily associated with PTSD symptom trajectory rather than these potential confounders. Future studies should systematically assess lifetime IPV exposure profiles (including cumulative duration and frequency), comorbid psychiatric conditions, and medication history to better disentangle these effects from those specific to PTSD persistence versus remission. While the remitted PTSD group comprised individuals who previously met full diagnostic criteria for PTSD but no longer met criteria at the time of assessment, detailed information regarding the timing of symptom remission (i.e. how recently or how many years ago remission occurred) was not available. This limits our ability to determine whether the intermediate gray matter volumes observed in this group reflect incomplete structural normalisation (in cases of more recent remission) or a relatively stable, recovered state (in cases of longer-term remission). Additionally, the study did not include a trauma-exposed control group who never developed PTSD (i.e. resilient trauma-exposed individuals). Consequently, it remains unclear whether the gray matter volume differences observed between the remitted PTSD group and healthy controls are primarily attributable to prior trauma exposure itself or to the historical presence and subsequent remission of PTSD symptoms. Inclusion of such a resilient trauma-exposed group in future studies would help disentangle these effects and strengthen causal interpretations. Furthermore, the distinction in statistical approaches for gray matter volume and inter-network connectivity arises from the inherent nature of structural covariance analysis: gray matter volume provides individual-level metrics amenable to regression-based models, whereas connectivity yields group-level summary coefficients requiring Z-tests for comparison. This limitation highlights the need for future studies to incorporate multimodal neuroimaging, such as diffusion tensor imaging (DTI), to derive individual-level metrics of structural connectivity and replicate our findings. Finally, structural covariance analysis is an indirect method that infers networks based on correlations in gray matter volume. It does not directly measure axonal connectivity, as assessed by diffusion tensor imaging, or functional co-activation, as evaluated with functional magnetic resonance imaging. Incorporating multimodal neuroimaging approaches in future work will enhance the precision and interpretability of these findings. In conclusion, this study provides novel evidence that persistent PTSD is characterised by distinct structural network alterations, including gray matter volume reductions that parallel clinical severity and heightened hyperconnectivity between cognitive control and limbic-interoceptive systems. Remitted PTSD exhibited connectivity patterns comparable to healthy controls, consistent with models emphasising the restoration of balanced network interactions during recovery. Together, these findings advance understanding of the neurobiological mechanisms underlying PTSD chronicity and may inform the development of neurobiologically-grounded strategies for understanding and managing for treatment-resistant cases. Supplementary Material Supplementary Material.docx Funding Statement This research was supported by the National Research Foundation of Korea funded by the Korean government (RS-2020-NR049587, RS-2024-00440371, RS-2024-00345975, and RS-2023-00249051). Disclosure statement No potential conflict of interest was reported by the author(s). Data availability statement The data that support the findings of this study are available from the corresponding authors upon reasonable request. The data are not publicly available owing to privacy and ethical restrictions. Supplemental Material Supplemental data for this article can be accessed online at https://doi.org/10.1080/20008066.2026.2655604. References Associated Data Supplementary Materials Supplementary Material.docx Data Availability Statement The data that support the findings of this study are available from the corresponding authors upon reasonable request. The data are not publicly available owing to privacy and ethical restrictions.