Widespread Cortical Thinning and Subcortical Alterations in Mood and Psychotic Disorders Across First-Episode and Chronic Stages

Abstract

Bipolar disorder (BP) and schizophrenia (SZ) share overlapping yet distinct structural brain alterations. Clarifying these differences is essential for improving diagnosis and prognosis. We hypothesized that cortical and subcortical alterations would differ between BP and SZ across illness stages and that these changes would relate to clinical severity. We examined cortical thickness and subcortical volumes using T1-weighted MRI in 438 participants recruited at McLean Hospital: 122 BP, 155 SZ, and 155 healthy controls (HC). Patient groups included both first-episode (FE-BP, FE-SZ) and chronic (CH-BP, CH-SZ) stages. BP patients demonstrated widespread cortical thinning, particularly in frontal, parietal, and temporal regions, with greater reductions in CH-BP. In CH-SZ, thinning was more circumscribed, affecting the precentral, superior frontal, and inferior parietal cortices. Subcortically, FE-BP showed corpus callosum (CC) volume loss, with further reductions in CH-BP, while CH-SZ exhibited central CC thinning. Ventricular enlargement was observed in CH-BP, FE-SZ, and CH-SZ, with marked left lateral ventricle expansion in CH-BP and CH-SZ. FE-BP also showed reduced caudate and amygdala volumes, with similar trends in chronic stages. Clinically, PANSS negative scores were consistently higher in SZ than BP, whereas positive symptoms did not differ. Cortical and subcortical alterations correlated with disease duration and PANSS scores, linking brain changes to symptom burden. BP exhibited larger morphometric effect sizes than SZ, despite greater negative symptom severity in SZ. This dissociation suggests that structural alterations do not necessarily parallel clinical severity, underscoring the importance of integrating neuroimaging and clinical measures to refine disease trajectories and diagnostic boundaries.

1. Introduction

Schizophrenia (SZ) is a primary psychotic disorder, whereas bipolar disorder (BP) is an affective disorder in which psychosis occurs only in a subset of cases; both conditions present substantial complexity and impose major demands on healthcare systems. Despite advancements, effective treatments remain elusive. BP is characterized by manic episodes, featuring elevated energy and euphoria, and depressive episodes, marked by low mood and energy. These episodes can severely impair daily functioning. BP affects about 1–2% of the global population, often manifesting in late adolescence or early adulthood¹. Schizophrenia is a chronic psychotic disorder affecting approximately 1% of the population². It is marked by positive symptoms such as hallucinations and delusions, negative symptoms like emotional blunting and social withdrawal, and cognitive deficits that impair memory and executive functioning³. BP and SZ likely result from a combination of neurochemical imbalances, particularly dysregulation of the glutamatergic⁴ and dopaminergic system⁵, genetic vulnerability and environmental factors, such as prenatal stress or infections⁶. They differ considerably in their symptoms, underlying mechanisms, and treatment approaches, but also share some overlapping common features in psychotic symptoms and biological mechanisms. During manic or depressive episodes, individuals with BP may experience delusions or hallucinations similar to those seen in SZ⁷. However, these symptoms are episodic in BP, tied to mood states, while they are more persistent and pervasive in SZ⁷. Another key difference lies in cognitive function. While both conditions involve some cognitive impairment, it is more pronounced and widespread in SZ⁸. Individuals with SZ often struggle with memory, attention, and problem-solving to a degree that affects their daily functioning more severely than in BP⁹.

Brain structural alterations are evident in both SZ and BP, reflecting distinct but overlapping neuropathological features. Enlargement of brain ventricles, particularly the lateral and third ventricles, has been consistently observed in SZ and BP^10–12 reflecting structural brain abnormalities associated with these psychotic conditions. Cortical thinning is also a shared neurobiological feature observed in both SZ and BP, with distinct and overlapping regional patterns^13–15. In SZ, significant cortical thinning has been identified in the frontal, temporal, and parietal regions, which are associated with cognitive impairments and symptom severity¹⁶. Similarly, in BP, cortical thinning is also observed in the frontal and temporal cortices^17,18. In both disorders, structural changes are often observed in regions involved in emotion regulation, such as the prefrontal cortex, amygdala, and hippocampus. Studies show reduced gray matter volume in the prefrontal cortex and abnormalities in the amygdala, which may contribute to mood dysregulation¹⁹. Gray matter reduction is closely associated with the onset of psychotic disorders, with structural changes often stabilizing in later stages^20,21. Prior research has consistently demonstrated that structural brain alterations differ between early and chronic stages of both SZ and BP. In SZ, early-stage patients often exhibit gray matter reductions in frontal and temporal regions, with more widespread and pronounced changes emerging with disease progression^11,22. Similarly, BP patients show cortical and subcortical abnormalities that may become more evident over time, particularly in those with a history of multiple mood episodes or psychosis^23,24. In addition to cortical and subcortical abnormalities, alterations in major white-matter pathways, particularly the corpus callosum (CC), are increasingly recognized in both SZ and BP^25–28. The CC is the largest interhemispheric fiber bundle and is essential for integrating emotional, cognitive, and sensory–motor information across hemispheres. Disruptions in CC microstructure have been linked to impaired cognitive control, mood dysregulation, and psychotic symptomatology, with diffusion MRI studies reporting reduced fractional anisotropy and altered callosal thickness in both SZ and BP^29–31.

While these findings support the progressive nature of structural changes in psychotic disorders, direct comparisons between SZ and BP across illness stages remain limited. Furthermore, the extent to which these abnormalities relate to clinical symptom severity and disease duration is still not well understood, underscoring the need for further investigation.

This study compared structural brain alterations in SZ and BP across first-episode (FE) and chronic (CH) stages using T1-weighted MRI. We analyzed cortical thickness, surface area, and volume, each reflecting distinct neurobiological processes: thickness (microstructural changes³²), area (neurodevelopment³³), and volume (global tissue integrity³⁴). We hypothesized that structural differences would be detectable across the five study groups (HC, FE-SZ, CH-SZ, FE-BP, CH-BP), with more pronounced abnormalities in chronic (CH) relative to first-episode (FE) patients. We additionally expected SZ subgroups to show greater cortical thinning than BP subgroups, and anticipated associations between structural measures, PANSS symptoms, and illness duration.

The unique contributions of this study include a direct cross-sectional comparison of first-episode and chronic groups within and between SZ and BP, with all participants recruited and clinically evaluated at the same institution using consistent diagnostic procedures and uniform image-processing pipelines. Although MRI data were acquired on two Siemens scanners, scanner-related variability was addressed by including scanner type as a covariate in all analyses, ensuring that observed effects were not driven by acquisition differences. Additional strengths include the simultaneous analysis of cortical thickness, surface area, and subcortical volume to clarify their distinct and overlapping contributions, and the incorporation of clinical correlations that extend across diagnostic boundaries. While previous studies have examined similar measures, our design provides clearer insights into stage-specific neuroanatomical changes and their clinical relevance.

2. Materials & Methods

2.1. Participants

Participant data were collected from the McLean Psychotic Disorders Division Clinical Data Repository. Patients were recruited from clinical services at the hospital, including McLean OnTrack, a specialty first-episode psychosis treatment service. Healthy individuals were recruited via advertisement. All participants were informed of the detailed description of the studies and given consent forms, approved by the Mass General Brigham Institutional Review Board. Diagnostic assessments were carried out by trained psychiatrists and clinical psychologists using the Structured Clinical Interview for DSM-IV for patients and HC. Symptom severity was measured using the Positive and Negative Syndrome Scale (PANSS), a widely used tool for assessing positive, negative, and general psychopathology in schizophrenia. The procedures of clinical assessments and recruitment criteria were described in our previous publications³⁵. In this study, patients with a duration of illness less than three years were classified as first-episode, and we note that individuals with less than one year of illness had been grouped similarly in our prior publications due to limited sample sizes.

469 participants’ data were analyzed in this study, include 61 FE-BP (35 males and 26 females; age 25 ±5), 73 CH-BP (40 males and 33 females; age 34 ±9), 59 FE-SZ (39 males and 20 females; age 24 ± 5),131 CH-SZ (90 males and 41 females; age 35 ± 9) and 145 HC (69 males and 76 females; age 29 ± 9). All bipolar patients included in this study meet criteria for Bipolar I disorder. Demographic information and clinical scores were shown in Table 1.

Table 1.

Demographic and clinical information of the patients and controls.

	FE-BP (N=61)	CH-BP (N=73)	FE-SZ (N=59)	CH-SZ (N=131)	HC (N=145)	P
Age (years)	25.00±5.45	34.16±9.49	24.38±5.07	34.77±9.38	28.84±9.06	1.96*e-20
Age range (years)	19–50	20–65	19–46	21–58	18–57
Sex (male/female)	35/26	40/33	39/20	90/41	69/76	0.00581
Disease duration (years)	1.38±0.87	12.59±8.48	1.73±1.04	13.86±8.52	-	<0.0012
Age of onset (years)	23.62±5.52	21.57±12.73	22.65±5.18	20.91±12.67	-	0.39
PANSS positive	16.92±9.02	14.95±7.60	16.34±6.83	17.43±6.98	-	0.152
PANSS negative	10.13±3.56	9.78±2.80	16.69±7.56	14.91±6.59	-	<0.0012
PANSS general	27.18±7.40	27.82±8.96	31.29±8.25	31.56±9.68	-	0.0042
Subjects: scanner 1 (n=)	10	10	27	52	74
Subjects: scanner 2 (n=)	51	63	32	79	71

¹Chi-square test

²Kruskal-Wallis

2.2. MRI data acquisition

Imaging data were acquired using two Siemens MRI scanners at Mclean Imaging Center. Scanner 1: 3-T Siemens MAGNETOM Prisma equipped with a 64-channel head coil. T1-weighted magnetization-prepared rapid acquisition having gradient multi-echo (MPRAGE) imaging was acquired with the following acquisition parameters: repetition time (TR)=2530 ms; echo time (TE)=1.69 ms; field of view (FOV)=256 mm; voxel dimensions =1.0×1.0×1.0 mm³; 176 slices. Scanner 2: 3-T Siemens Tim-Trio equipped with a 32-channel phased-array head coil. T1- MPRAGE imaging was collected with the following parameters: TR=1900 ms; TE=2.26 ms; FOV=250 mm; voxel dimensions =1.0×1.0×1.0 mm³; 176 slices.

2.3. Image preprocessing

All T1-images were inspected for quality and motion artifacts. Processing was done using the FreeSurfer v. 7.2.0 (http://surfer.nmr.mgh.harvard.edu [Boston, USA]) recon-all pipeline with the default settings. Segmented labels were visually inspected to identify potential segmentation artifacts and manual corrections were applied when needed. Segmentations were discarded from the study if manual corrections were not possible, ultimately all the participants were included for the cortical analyses, while 2 FE-SZ, 1 CH-SZ and 1 FE-BP were excluded for the subcortical analyses. The cortical thickness of each vertex of the participants’ surfaces was mapped on a common spherical coordinate system using spherical transformation. The registered cortical thickness of surface vertices was smoothed using a Gaussian kernel of 10 mm. The cortical area and volume also were inspected, following the similar procedures. The FreeSurfer output of subcortical volumes (caudate, putamen, globus pallidus, nucleus accumbens, amygdala, hippocampus, thalamus), corpus callosum, and global volumes such as the total gray and cerebral white matter, were used as feature groups in the analysis. All subcortical volumetric measures were normalized to estimated total intracranial volume (eTIV) to account for individual differences in head size. eTIV was derived from FreeSurfer’s automated segmentation output. Normalization was performed using the proportion method, in which each raw regional volume was divided by the participant’s eTIV, producing eTIV-adjusted volumetric measures for subsequent statistical analyses.

2.4. Statistical analysis

Clinical Measures Analysis

To examine group differences in clinical measures, we compared disease duration and PANSS symptom scores (positive, negative, and general subscales) across diagnostic groups: CH-BP (n = 73), FE-BP (n = 60), FE-SZ (n = 59), and CH-SZ (n = 131). Group comparisons were conducted using one-way analysis of variance (ANOVA) with diagnosis as the between-subjects factor. Post-hoc pairwise comparisons between diagnostic groups were performed using Bonferroni-corrected t-tests. Statistical significance was set at p < 0.05 (two-tailed).

Cortical Analysis

To examine group differences in cortical morphology, vertex-wise analyses of cortical thickness, surface area, and volume were conducted using FreeSurfer’s general linear model (GLM) framework. A single linear model was used for each morphological metric, with diagnostic group (HC, FE-BP, FE-SZ, CH-BP, CH-SZ) included as a between-subject factor, and age, sex, and scanner type included as covariates. This model allowed for a unified assessment of all group differences within one statistical framework. Cluster-wise correction for multiple comparisons was performed using FreeSurfer’s built-in Monte Carlo simulation procedure. Specifically: A cluster-forming threshold of p < 0.001 (two-tailed) was applied to the vertex-wise GLM results. Cluster-wise p-values (CWP) were calculated using 5000 permutations, and only clusters with CWP < 0.05 were considered statistically significant. Next, the resulting significant clusters were defined as regions of interest (ROIs) for follow-up correlation analyses with clinical variables. To assess the relationship between cortical morphology and clinical symptom severity, we performed correlation analyses between cortical thickness and PANSS scores/disease duration using FreeSurfer’s GLM DODS (Different Onset, Different Slope) modeling approach. All models included age, sex, scanner type, and eTIV as covariates to control for potential confounding influences. These analyses were restricted to patient groups and to regions where significant group-level morphological differences were identified in the vertex-wise analysis.

Subcortical Analysis

Group differences in global brain and subcortical volumes were assessed using analysis of covariance (ANCOVA), with age, sex, and scanner type included as covariates. For each region of interest, we first tested for a main effect of diagnostic group (HC, FE-BP, FE-SZ, CH-BP, CH-SZ). When a significant main effect was observed, we conducted all possible pairwise group comparisons. Post hoc group comparisons were corrected for multiple testing within each ANOVA using the Šidák adjustment. To account for multiple comparisons across all subcortical regions, we further applied the Benjamini–Hochberg False Discovery Rate (FDR) procedure, with significance defined as q < 0.05. For transparency, we report the test statistic (F or t), uncorrected p-value, FDR-corrected q-value, and effect size (Cohen’s d) for each contrast. Subcortical structures, including the corpus callosum, were segmented using FreeSurfer’s automated subcortical segmentation pipeline (aseg.mgz), which partitions the corpus callosum into five anatomically defined subregions. Associations between subcortical volumes and behavioral measures (e.g., PANSS scores, disease duration) were assessed using partial Pearson correlation analyses, controlling for age, sex, scanner type, and eTIV to ensure that observed relationships were not confounded by demographic or scanner-related factors. FDR correction was applied to multiple tests within each set of correlations. All analyses were performed using GraphPad Prism (version 9.0.0) and Statistica (version 14.0.1.25).

Results

3.1. Quality assessment between two-scanners

Because participants were scanned on two different MRI systems (Scanner 1: n=173; Scanner 2: n=196), we performed a targeted quality assessment to evaluate whether basic image-quality metrics differed across scanners. We selected healthy control (HC) males (Scanner 1: n = 35; Scanner 2: n = 34) for this comparison because (1) this was the largest demographic subgroup scanned on both systems, ensuring stable estimates, and (2) restricting analyses to a single sex reduces variance associated with known sex-related differences in brain morphology. The two scanner subgroups did not significantly differ in age (p > 0.20).

Quality metrics showed no significant scanner effects in white-matter SNR (p > 0.559), gray-matter SNR (p > 0.672), total gray matter volume (p > 0.141), cortical white matter volume (p > 0.383), CSF volume (p > 0.200), or eTIV (p > 0.322). These results support comparable acquisition quality across scanners for this matched subgroup. Because scanner effects could still influence other groups, scanner type was included as a covariate in all statistical analyses. (Supplementary Figure 1).

3.2. Clinical measures findings

ANOVA revealed significant group differences in disease duration (F(3,317) = 63.87, p < 0.001), PANSS Negative scores (F(3,317) = 27.91, p < 0.001), and PANSS General scores (F(3,317) = 4.90, p = 0.002), while PANSS Positive scores did not differ significantly (F(3,317) = 1.99, p = 0. 12). Disease duration: As expected, chronic patients had markedly longer illness durations (CH-BP: 12.38 ± 0.98 years; CH-SZ: 13.91 ± 0.77 years) compared with first-episode groups (FE-BP: 1.61 ± 0.20 years; FE-SZ: 2.51 ± 0.49 years) (all p < 0.001). CH-BP and CH-SZ did not differ significantly, nor did FE-BP and FE-SZ. PANSS Negative: FE-SZ (16.46 ± 0.91) and CH-SZ (14.94 ± 0.55) showed significantly higher negative symptom burden than CH-BP (9.92 ± 0.32) and FE-BP (10.08 ± 0.46) (all p < 0.01), whereas CH-BP and FE-BP did not differ. PANSS General: FE-SZ (31.09 ± 1.11) and CH-SZ (31.69 ± 0.85) scored higher than CH-BP (28.11 ± 1.02) and FE-BP (27.22 ± 0.99) (p < 0.05). No significant differences were found between FE-SZ and CH-SZ, or between CH-BP and FE-BP. PANSS Positive: No group differences were detected (CH-BP: 14.93 ± 0.87; FE-BP: 16.75 ± 1.16; FE_SZ: 15.89 ± 0.89; CH-SZ: 17.49 ± 0.62) (Figure 1). These results demonstrate that disease duration distinguishes chronic from first-episode patients, while negative and general symptoms are elevated in schizophrenia (both first-episode and chronic). In contrast, positive symptoms did not significantly differ across groups.

Figure 1.

Disease duration and PANSS scores across diagnostic groups. Violin plots with group means ± SEM show longer illness duration in chronic vs. first-episode groups and higher negative symptoms in SZ vs. BP, while positive or general scores did not differ. *p < 0.05, **p < 0.01.

3.3. Cortical findings

We observed reduced cortical thickness in both the FE-BP and CH-BP groups compared to HC. In the FE-BP group, affected regions included the left- pars opercularis, inferior parietal, supramarginal, and superior temporal areas, as well as the right- supramarginal, superior temporal, and precuneus regions (Figure 2A, and Supplementary Table 1). In the CH-BP group, reduced thickness was observed in the left- precentral, superior parietal, lateral occipital, superior frontal, transverse temporal, rostral middle frontal, precuneus, and pericalcarine regions, as well as in the right- caudal middle frontal, rostral middle frontal, pericalcarine, superior frontal, superior temporal, rostral anterior cingulate, pars triangularis, precentral, and supramarginal regions (Figure 2B, and Supplementary Table 1). No significant differences in cortical thickness were observed between the FE-SZ group and the HC group (Figure 2C). However, in the CH-SZ group, thinner cortical regions were identified. These regions included the left- precentral, pars triangularis, supramarginal, superior frontal, rostral middle frontal, and postcentral areas as well as the right- rostral middle frontal, superior temporal, inferior parietal, and superior frontal regions (Figure 2D, and Supplementary Table 1).

Figure 2.

Thickness differences among patient groups and healthy controls. Brain regions with significantly reduced thickness (indicated in red) in FE-BP (A), CH-BP (B), FE-SZ (C) and CH-SZ (D) compared to HC, as well as FE-BP (E) compared to FE-SZ and CH-BP (F) compared to CH-SZ.

Additionally, FE-BP group showed reduced thickness in the transverse temporal region compared to the FE-SZ group (Figure 2E). Likewise, the CH-BP group exhibited decreased thickness in the supramarginal region compared to the CH-SZ group (Figure 2F). Notably, both regions also showed reduced thickness relative to the HC.

In addition to cortical thickness, we also observed differences in volume and area between the patient groups and HCs. Significantly reduced volume differences were found in the CH-BP group, particularly in the left lateral orbitofrontal, pars opercularis, insula, superior frontal regions, and the right lingual region. In the CH-SZ group, significant reduced volume differences were observed in the left lateral orbitofrontal, pars opercularis, lingual, and right superior frontal, pars opercularis, and medial orbitofrontal regions (Supplementary Figure 2). The CH-BP group also showed a significant reduction in surface area of the pars opercularis. No significant differences were found between patient groups in volume or area measurements. All statistically significant regions are listed in Supplementary Table 2.

Our findings demonstrate modest but regionally consistent cortical thickness reductions across patient groups, particularly in frontal and temporal regions. Notably, cortical thinning was observed in FE-BP but not in FE-SZ when compared to healthy controls, while CH-SZ showed more widespread reductions. Finally, individuals with bipolar disorder showed more pronounced reductions in supramarginal and transverse temporal regions compared to those with schizophrenia.

3.4. Correlation between cortical thickness, disease duration, PANSS scores.

In addition to the above identified cortex thickness clusters showing differences compared to HC, we also examined the association between the affected thickness regions and clinical attributes in the patient groups. The detailed results were presented in the Supplementary Tables 1-2. Representatively, there exist some strong correlations in the BP group (Figure 3). In the FE-BP group, cortical thickness in the left supramarginal cluster demonstrated a significantly negative correlation with PANSS positive scores (r = −0.43, p = 5 × 10⁻⁴, q = 0.0020). In the CH-BP group, the left superior parietal thickness negatively correlated with PANSS positive scores (r = −0.34, p = 0.003, q = 0.0080), while the cortical thickness of left pericalcarine cluster positively correlated with PANSS general scores (r = 0.44, p = 8.94 × 10⁻⁵, = 0.00054). In the right hemisphere, the rostral middle frontal region positively correlated with both disease duration (r = 0.48, p = 1.76 × 10⁻⁵, q = 0.00021) and PANSS positive scores (r = 0.28, p = 0.017, q = 0.0216).

Figure 3.

Representative correlation plots for first-episode and chronic bipolar disorder. The regression and 95% confidence intervals for the regression were presented by the black solid line and the gray shaded area.

In the left hemisphere, we found that the thickness of a cluster in the superior frontal region positively correlated with PANSS positive (r = 0.19, p = 0.027, q = 0.0295) and PANSS general (r = 0.25, p = 0.004, q = 0.0080) scores in the CH-SZ group. Additionally, in the same hemisphere, cortical thickness of a cluster in the rostral middle frontal region positively correlated with disease duration (r = 0.19, p = 0.032, q = 0.0320) and PANSS general (r = 0.21, p = 0.018, q = 0.0216). Similar positive correlations with disease duration were observed in the inferior parietal region (r = 0.22, p = 0.013, q = 0.0195), and in the superior frontal region for PANSS positive (r = 0.22, p = 0.012, q = 0.0195) and PANSS general (r = 0.25, p = 0.004, q = 0.0080).

3.5. Subcortical Findings

Compared to healthy controls (HC), disorder-specific volumetric alterations were observed across patient groups (Figure 4). The third ventricle volume was significantly larger in chronic bipolar disorder (CH-BP; p = 0.0005, q = 0.0020, Cohen’s d = 0.60), first-episode schizophrenia (FE-SZ; p = 0.0252, q = 0.0279, Cohen’s d = 0.40), and chronic schizophrenia (CH-SZ; p < 0.0001, q = 0.0008, Cohen’s d = 0.57). Similarly, left lateral ventricle volume was increased in CH-BP (p = 0.0011, q = 0.0029, Cohen’s d = 0.47) and CH-SZ (p = 0.0026, q = 0.0050, Cohen’s d = 0.43). With respect to subcortical gray matter structures, the FE-BP group exhibited significantly reduced volumes in the left caudate (p = 0.0279, q = 0.0279, Cohen’s d = 0.44) and left amygdala (p = 0.0031, q = 0.0050, Cohen’s d = 0.50). In the CH-BP group, these same structures showed a trend toward reduced volumes relative to HC. Additionally, left amygdala volume was significantly lower in CH-SZ compared to HC (p = 0.0104, q = 0.0139, Cohen’s d = 0.30).

Figure 4.

Mean ± standard deviation for eTIV-normalized volumetric measures of subcortical structures in patient groups and healthy control groups and patient group comparisons. Statistical significance (uncorrected p) of group differences is indicated numerically.

As part of the overall ANOVA including all five groups (HC, FE-BP, FE-SZ, CH-BP, and CH-SZ), we observed a significant main effect of group for left amygdala volume (F(4, N–5) = 2.195, p < 0.0001). Post-hoc pairwise testing with Šidák correction revealed that the FE-BP group showed reduced left amygdala volume relative to the FE-SZ group (t(124) = 3.134, p = 0.0022, q = 0.0486, Cohen’s d = 0.47). Similarly, a main effect of group was observed for left thalamus volume (F(4, N–5) = 4.072, p = 0.0022). Post-hoc Šidák comparisons indicated greater volume in CH-SZ compared to CH-BP (t(308) = 2.704, p = 0.0072, q = 0.0626, Cohen’s d = 0.31), although this difference did not survive correction for multiple comparisons. Together, these analyses highlight both shared and diagnosis-specific alterations across the psychosis–bipolar spectrum. For transparency, we report both uncorrected and FDR-corrected p-values, as well as effect sizes.

3.6. Corpus Callosum

Significant differences in the corpus callosum (CC) were primarily observed in the bipolar disorder (BP) groups compared to healthy controls (HC) (Figure 5). Whole CC volume was reduced in FE-BP (p = 0.019, q = 0.019, Cohen’s d = 0.47) and CH-BP (p = 0.010, q = 0.012, Cohen’s d = 0.45). Subregional analyses revealed marked reductions in the mid-posterior CC in CH-BP (p < 0.0001, q = 0.00035, Cohen’s d = 0.62) and in the central CC for both CH-BP (p < 0.0001, q = 0.00035, Cohen’s d = 0.63) and CH-SZ (p = 0.0004, q = 0.00093, Cohen’s d = 0.45). Furthermore, the mid-anterior CC was significantly reduced in CH-BP (p = 0.0010, q = 0.0014, Cohen’s d = 0.55), and the anterior CC was reduced in FE-BP (p = 0.0007, q = 0.0012, Cohen’s d = 0.56).

Figure 5.

Mean ± standard deviation for eTIV-normalized corpus callosum segmentation volumes in patient groups and healthy control groups. Statistical significance (uncorrected p) of group differences is indicated numerically.

The CC volumes between FE-SZ and FE-BP groups showed no significant differences. However, in the chronic groups, multiple volume differences emerged. Specifically, the CH-BP group showed reduced corpus callosum volumes compared to CH-SZ in the posterior (t(308) = 2.951, p = 0.0034, q = 0.0201, Cohen’s d = 0.41), mid-posterior (t(308) = 2.599, p = 0.0098, q = 0.0255, Cohen’s d = 0.32), mid-anterior (t(308) = 2.712, p = 0.0071, q = 0.0307, Cohen’s d = 0.32), and anterior (t(308) = 2.668, p = 0.0080, q = 0.0253, Cohen’s d = 0.31) subregions (Supplementary Figure 3).

3.7. Correlation between subcortical volumes, disease duration, PANSS scores

In CH-BP, a positive correlation was observed between disease duration and the volumes of the left lateral ventricle (r = 0.34, p = 0.0028, q = 0.0042) and the third ventricle (r = 0.58, p < 0.0001, q = 0.00030). Additionally, the volume of the anterior CC was positively correlated with the PANSS general score (r = 0.36, p = 0.0044, q = 0.0044) (Figure 6).

Figure 6.

Correlation plots in chronic bipolar disorder (BP). Black solid-line: regression-line; gray shaded area indicates the 95% confidence intervals for regression lines.

4. Discussion

This study examined cortical and subcortical morphological changes in FE and CH BP, SZ, and HC groups from a single center, alongside associations with illness duration and clinical scores. Key findings include more pronounced structural alterations in BP groups compared to HC, with weaker effects in SZ. Furthermore, we found that negative and general PANSS scores were elevated in schizophrenia relative to bipolar disorder, whereas positive symptom scores did not differ across groups. Chronic groups also showed longer illness duration compared with first-episode patients, as expected. However, correlations between brain measures and symptom severity (PANSS) were inconsistent in directionality across regions, raising questions about their utility in characterizing disease-specific pathways.

4.1. Cortical Changes in BP: Replication and Divergence

Both FE-BP and CH-BP showed reduced cortical thickness and volume relative to HC, most prominently in prefrontal (middle frontal), temporal (superior temporal), and parietal regions (inferior parietal lobule, supramarginal gyrus), consistent with prior literature^14,18. These findings are also broadly consistent with previous studies linking structural alterations in these regions to mood regulation and social cognition circuits^36–38. Although CH-BP showed a somewhat broader spatial distribution of reductions, direct comparisons between FE-BP and CH-BP did not reveal statistically significant differences. Thus, while our results align with prior reports of structural alterations in bipolar disorder, they do not provide conclusive evidence for progressive cortical differences between first-episode and chronic groups. Interestingly, our BP groups showed larger effect sizes than SZ, a pattern that contrasts with reports of more severe SZ-related thinning in some consortia analyses^11,39,40. This discrepancy may reflect cohort-specific factors such as medication exposure, illness subtype heterogeneity, or different neurodevelopmental trajectories across diagnoses. The absence of significant cortical thinning in FE-SZ (in contrast to FE-BP) also raises questions about the timing and detectability of early structural markers in SZ

Large-scale multi-site studies have reported widespread cortical thinning across both SZ and BP relative to HC, with robust effects across many regions, including fusiform cortex as a site of strongest reductions⁴⁰. In contrast, our single-site cohort revealed fewer significant clusters, but in directionally consistent regions. Together, these results underscore the value of examining cortical metrics across illness stages within the same site, where harmonized acquisition and analytic pipelines can minimize methodological variability.

4.2. PANSS Correlations: Inconsistencies and Interpretation

We found that negative and general PANSS scores were elevated in schizophrenia compared with bipolar disorder, whereas positive symptoms did not differ significantly across groups, consistent with prior reports indicating that positive symptoms are less reliable for distinguishing between the two disorders⁴¹. These findings provide important clinical context for the brain morphometric analyses. Despite the greater negative symptom burden in SZ, we observed larger effect sizes in brain morphometric alterations in BP compared to SZ. This pattern contrasts with much of the extant literature, where SZ typically demonstrates more widespread deficits^11,23,40. One potential explanation is the contribution of lithium treatment, which has been associated with increased cortical and subcortical volumes in BP^37,42, possibly exaggerating structural group differences relative to SZ. In addition, inter-site variability in multi-site consortia may attenuate effect sizes for SZ^11,43, whereas more homogeneous sampling may preserve effects in BP cohorts.

In this study, the associations between cortical metrics and PANSS scores varied by group and direction. In FE-BP, thinner left supramarginal and superior temporal regions were correlated with higher PANSS positive and general scores. In CH-BP, reduced left superior parietal thickness was linked to higher PANSS positive scores, while thinning in the left pericalcarine region was associated with longer illness duration. In CH-SZ, prefrontal thinning showed modest associations with PANSS scores, although no group-level differences were observed compared to healthy controls.

The instability of these correlations, where thinner cortex predicts worse symptoms in some regions but not others, suggests PANSS may not reliably map onto specific structural changes. This mirrors debates about whether symptom scales reflect distinct neurobiological pathways or heterogeneous clinical constructs⁴⁴. Future work should integrate multimodal data (e.g., cognitive testing, functional imaging) to clarify these relationships.

4.3. Subcortical Findings: BP vs. SZ Trajectories

Amygdala volume reductions in FE-BP (but not CH-BP) and corpus callosum (CC) thinning in CH-BP suggest early limbic involvement with later callosal progression, possibly tied to mood dysregulation^45–47. Third ventricle enlargement in both FE-SZ and CH-SZ (more pronounced in chronic stages) aligns with neurodegenerative hypotheses^48,49, while thalamic volume increases in CH-SZ vs. CH-BP highlight diagnostic-divergent pathways⁵⁰.

While prior studies have frequently reported reduced volumes in the hippocampus and thalamus in chronic- SZ and BP disorder^12,51–54, our findings did not fully replicate these patterns, particularly in the chronic groups. One possible explanation is that our cohort was recruited from a single site (McLean Hospital), which may reduce inter-site variability but may also limit generalizability. Moreover, several clinical and demographic variables, such as medication history, symptom severity, or number of mood/psychotic episodes were not directly controlled in our analysis and may have influenced subcortical volumes. It is also possible that some subcortical changes, particularly in the hippocampus and thalamus, may emerge more clearly in longitudinal studies. In the CH-BP group, we observed structural associations that may inform understanding of disease progression, though these findings require cautious interpretation. While ventricular volumes showed positive correlations with illness duration, supporting previous reports of progressive enlargement in BP¹², the directionality of CC associations was less consistent. The anterior CC volume correlated positively with PANSS general scores, while mid-posterior CC thickness showed negative correlations with symptom severity. These mixed findings mirror the broader pattern of region-specific associations seen throughout our study, where structural alterations related to symptoms in opposing directions depending on the brain region examined. Such inconsistencies suggest that: (1) PANSS scores may capture multiple biological processes with distinct neuroanatomical substrates, and/or (2) the relationship between structure and symptomatology in BP may be non-linear or moderated by unmeasured variables (e.g., treatment effects, illness subtypes). While the anterior CC’s role in interhemispheric communication is well-established^55–57, our data emphasize the need for more nuanced models that account for both localized structural changes and their potentially divergent clinical correlates in BP.

While our results reinforce cortical thinning and subcortical volume changes as neuroanatomical features of both BP and SZ, several critical nuances warrant emphasis. First, the finding that BP groups showed larger effect sizes than SZ in structural alterations contrasts with much of the extant literature, where SZ typically demonstrates more widespread deficits. This discrepancy may reflect: Cohort-specific factors, such as differences in illness severity, medication exposure, or comorbid conditions in our sample. Methodological considerations, including cross-sectional design and potential partial volume effects in subcortical segmentation, which may differentially impact BP and SZ measurements. Temporal dynamics: While SZ is often characterized by progressive gray matter loss, BP may show more pronounced early structural changes that stabilize or become less distinguishable from SZ in chronic stages.

Second, the instability of PANSS correlations, where thinner cortex paradoxically associated with both higher and lower symptom severity depending on the region, highlights fundamental challenges in linking structural measures to clinical phenomenology. These inconsistencies suggest that: PANSS scores may represent heterogeneous constructs with distinct neurobiological substrates (e.g., positive symptoms could relate to temporal lobe structure, while negative symptoms may reflect prefrontal integrity). Non-linear relationships may exist (e.g., U-shaped effects where both excessive thinning and preserved thickness correlate with poor outcomes in different patient subsets). Unaccounted confounders (e.g., medication effects, illness episodes, or cognitive reserve) could obscure straightforward structure-function mappings.

Moving forward, we propose three key priorities for future research: Illness Stage Stratification: Our FE versus CH group differences suggest dynamic neurostructural trajectories. Longitudinal designs tracking individuals from prodrome to chronic stages could clarify whether BP and SZ diverge early but converge later. Treatment-Effect Disentanglement: Given evidence that lithium may increase cortical thickness in BP⁵⁸ while antipsychotics may accelerate volume loss in SZ⁵⁹, future studies should integrate detailed medication histories and consider prospective medication-naïve designs. Symptom-Specific Modeling: Rather than using PANSS total scores, dimensional approaches (e.g., biotypes or symptom clusters) may better capture biologically meaningful relationships^60,61.

4.4. Limitations

The main limitations of the study include the relatively modest sample sizes, which may have reduced statistical power to detect subtle brain differences and limited our ability to fully characterize stage-specific signatures of disease progression. All our data were collected from multiple projects, which prevented meaningful age-/sex-matched subgroup-analyses. Therefore, we included age and sex as covariates in all statistical models. Incomplete data on lifetime antipsychotic exposure also prevented a comprehensive assessment of how these medications may have influenced brain structure, making it difficult to disentangle illness-related neural changes from treatment effects. Another key limitation is the absence of standardized cognitive assessments (e.g., IQ or MATRICS). Although cognitive impairments, particularly in executive function and memory, are well documented in schizophrenia and bipolar disorder and are linked to brain structure, our analyses could not incorporate these associations due to limited and inconsistent cognitive data across participants. While the data were collected at a single site, they were derived from multiple studies conducted within that site, and not all included cognitive testing. Consequently, the available data were insufficient to examine potential group differences in cognitive ability or their relationship to cortical and subcortical alterations. Similarly, systematic data on substance use were not available. Substance use, particularly cannabis, alcohol, and stimulants, can impact brain structure and may partially confound observed group differences. Although participants were clinically assessed and acute intoxication or dependence was typically an exclusion criterion, the absence of detailed histories limited our ability to control for these effects. Importantly, although this was a single-site study, imaging data were acquired on two 3T Siemens MRI scanners (MAGNETOM Prisma and Tim-Trio). To account for scanner-related variance, scanner type was included as a covariate in all analyses. While this statistical approach helps mitigate potential confounds, we acknowledge that residual scanner effects cannot be fully excluded. Lastly, the cross-sectional design precludes causal inference regarding the relationship between brain structure and symptom severity. Longitudinal studies are needed to clarify how cortical and subcortical alterations evolve with illness course and treatment across diagnostic groups.

Replication and Future Directions

While large-scale multi-site studies such as ENIGMA have provided invaluable insights into structural abnormalities in SZ and BP, our study offers complementary strengths that are more difficult to achieve in large heterogeneous datasets. In particular, single-institution recruitment, uniform diagnostic procedures, and consistent imaging-processing pipelines allow for more precise comparisons between first-episode and chronic groups within and across diagnoses, analyses that are difficult to standardize across multi-site consortia. Our design reduces variability related to site-specific acquisition differences and clinical assessment protocols, thereby clarifying stage-related group differences that may be obscured in aggregated datasets.

Rather than emphasizing the need for replication solely through larger samples, the value of future work lies in independent cross-sectional cohorts with similarly controlled acquisition and harmonized pipelines, which can help determine which structural patterns are robust across settings and which are sample- or site-specific. Together, these complementary approaches, large-scale consortia and tightly controlled single-institution studies are essential for building a reproducible and mechanistically meaningful understanding of structural alterations in psychotic and affective disorders.

Supplementary Material

1

Supplementary Figure 1. There were no significant differences in global mean cortical thickness across the experimental groups. Statistical analysis using Tukey’s multiple comparisons test revealed that all pairwise group comparisons resulted in p-values > 0.07, indicating no significant differences between any of the groups.

Supplementary Figure 2. Volume and area differences between patient groups and healthy controls. Brain regions with reduced volume in CH-BP (A) and CH-SZ (B), and reduced area in CH-BP (C), compared to HC.

Supplementary Table 1. Results of thickness differences between patients’ groups and healthy controls, and correlations with clinical scores.

Supplementary Table 2. Results of volume and area differences between patients’ groups and healthy controls, and correlations with clinical scores.

Highlights.

• Bipolar patients show widespread cortical thinning; Schizophrenia patients show more localized thinning.

• Chronic bipolar and schizophrenia patients show distinct patterns of corpus callosum alteration.

• Ventricular enlargement is common in chronic bipolar-, schizophrenia- and first-episode schizophrenia.

• Brain changes correlate with illness duration and symptom severity in bipolar disorder.

• Structural change and clinical severity are dissociated in bipolar and schizophrenia.