Cortical morphometric inverse divergence alterations in first-episode, treatment-naïve COVID-19 secondary depression correlate with transcriptional signatures

Abstract

Background

Theneurobiological mechanisms underlying COVID-19 secondary depression(CSD) remain poorly understood, particularly regardingmacrostructural brain alterations and their transcriptionalcorrelations. This study aimed to investigate regional morphometricinverse divergence (MIND) alterations and their molecularassociations with cortical transcriptional signatures in CSDpatients.

Methods

80 first-episode, treatment-naïve CSD patients and 40 matched healthycontrols underwent high-resolution structural MRI. MIND networks wereconstructed across 308 cortical regions using five morphometricfeatures. Case-control differences were assessed using general linearmodels adjusting for age, sex, and intracranial volume. Partial leastsquares (PLS) regression was employed to link regional MINDalterations with cortical gene expression profiles, meta-analyticcognitive domains, and transcriptomic features includingcross-disorder similarity, functional pathways, cell-typespecificity, cortical layer distribution, and developmental timing.

Results

CSDpatients exhibited significantly elevated MIND values in cingulateand supramarginal regions associated with emotional, memory, andlanguage processing. These changes correlated with stress andcognitive symptoms, independent of infection frequency. PLS4-weightedgenes explained 17.7% of MIND variance and were enriched inneurodevelopmental and metabolic (PLS4+), as well as immune-related(PLS4-) pathways. PLS4- genes were linked to microglia, astrocytes,and cortical layer-I, while PLS4+ genes localized to layer-V.Developmental enrichment indicated disruptions during early (fetal,infant) and late (adult) stages.

Conclusions

This study reveals that regional MIND alterations in CSD are underpinned by distinct transcriptional signatures, highlighting neurodevelopmental and immune-related mechanisms. These findings offer a multiscale framework connecting brain morphometry with transcriptional architecture in post-COVID depression.

Clinical trials

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07544-2.

Introduction

The global coronavirus disease 2019 (COVID-19) pandemic has profoundly impacted not only physical health but also psychological well-being [1]. Post-COVID-19 syndrome, defined by persistent symptoms beyond the acute phase, is increasingly associated with psychiatric conditions, particularly anxiety and depression. These patients present with manifestations including insomnia, anxiety, depressive symptoms, and post-traumatic stress, affecting approximately 30–40% of individuals during or following infection, often persisting for more than 12 weeks [2]. In this study, post-COVID-19 patients with persisting depression were defined as COVID-19 secondary depression (CSD) individuals, referring to one subtype of “Long COVID” defined by World Health Organization (WHO) [3]. Despite significant advances in characterizing its clinical presentations, the underlying neurobiological mechanisms, particularly for the relationship between brain structural alterations and CSDs pathophysiology, remain largely unclear and necessitate further investigation.

Several neuroimaging researches have identified structural brain abnormalities in patients with major depressive disorder (MDD), including alterations in cortical thickness, white matter integrity, and gray matter volume [4, 5]. However, relying on isolated imaging metrics poses limitations in elucidating the complex interregional structural relationships underlying the disorder. Consequently, recent studies have shifted focus toward structural connectome disruptions derived from diffusion-weighted imaging (DWI) tractography, which have highlighted aberrant connectivity patterns, particularly within fronto-limbic circuits encompassing the anterior cingulate, posterior cingulate, and dorsolateral prefrontal cortexes [6, 7]. Although numerous investigations have linked individual morphological changes to both symptom severity and cognitive impairment in depression, the dysconnectivity hypothesis in CSD remains inadequately explored, partly due to inherent methodological challenges in accurately characterizing brain-wide anatomical networks [8].

Morphometric similarity analysis has recently gained prominence as a robust approach for characterizing coordinated brain morphometric patterns in aging and neuropsychiatric conditions [9, 10]. It quantifies the statistical interdependence of local anatomical features by integrating multiple morphometric measures into biologically informative connectomes. Compared to conventional brain structural connectomes, morphometric similarity network (MSN) provides superior robustness, reproducibility, and interpretability. In MSN, brain regions are represented as vectors of macrostructural attributes, with similarity estimated through pairwise correlations of normalized feature profiles [11]. Alterations in MSN have been reported linked to various psychiatric illnesses, such as MDD [12], bipolar disorder (BD) [13], and schizophrenia [14], showing associations with distinct transcriptional signatures. Nonetheless, MSN is limited by their reliance on regionally averaged and normalized feature statistics. To overcome these weaknesses, the Morphometric Inverse Divergence (MIND) framework was recently developed, capturing morphometric similarity through divergence measures across multiple features at the cortical vertex level. MIND networks exhibit greater stability, enhanced correspondence with cortical cytoarchitecture, stronger alignment with regional gene co-expression, and increased heritability compared to MSN [15]. Despite these advantages, the application of MIND in clinical populations remains scarce. To date, no studies have systematically examined MIND network alterations in individuals with CSD, leaving its neuropathological relevance unexamined.

Accumulating evidence highlights the critical role of genetic factors in shaping human brain networks, particularly in establishing metabolically demanding and functionally essential connections [16]. Hub brain regions, areas with high connectivity, often exhibit elevated transcriptional activity of genes involved in dendritic growth and synaptic development [17]. The comprehensive whole-brain transcriptional datasets, obtained from the Allen Human Brain Atlas (AHBA) database, have enabled researchers to bridge molecular-level disease-related transcriptional patterns with macroscale brain alterations across various psychiatric conditions [18]. However, an integrated assessment of MIND alterations and regional transcriptional pattern for CSD patients remains unestablished. Further researches are necessary to elucidate the molecular underpinnings of CSD pathogenesis and to identify potential therapeutic targets.

This study aimed to elucidate the molecular underpinnings of CSD by linking MIND network alterations to cortical transcriptional features. We first employed a novel cortical similarity-based method to calculate regional MIND values and identify CSD-specific deviations relative to healthy controls (HCs). To contextualize these changes, we investigated their associations with disease severity and cognitive domains derived from the Neurosynth meta-analytic framework. Subsequently, the partial least squares (PLS) regression was applied to relate CSD-associated MIND alterations to regional gene expression patterns using datasets from the AHBA. Permutation analyses assessed the statistical significance of gene overlaps between CSD-related transcriptional profiles and dysregulated genes identified in other psychiatric conditions. Functional characterization was performed using Metascape enrichment and protein-protein interaction (PPI) analyses to identify key molecular pathways. Additionally, cell-type specificity was examined by mapping implicated genes to canonical brain cell populations. Finally, enrichment analyses across cortical layers and developmental time windows were performed to examine the spatial and temporal dynamics of transcriptional signatures linked to CSD-related MIND disruptions. Collectively, our study offers novel insights into the pathophysiological landscape of CSD, revealing intricate links between macroscale brain morphology and specific molecular architecture.

Methods

Participants enrollment

A total of 120 participants aged from 18 to 70 years were enrolled in this study, including 80 first-episode, drug-naïve CSD patients and 40 HCs. CSD patients was defined as a subtype of “Long COVID”, consistent with the WHO criteria [3]. Specifically, CSD referred to patients with a continued presence or new onset of depressive symptoms at least three months after confirmed COVID-19 infection, persisting for a minimum of two months and not attributable to other neurological disorders. All COVID-19 infections were identified based on the participants’ self-reports and subsequently verified by nucleic acid PCR testing using two nasopharyngeal swabs conducted in our hospital. These patients were prospectively recruited from The Affiliated Yueqing Hospital of Wenzhou Medical University between January 2022 and January 2023. During the same period, age- and sex-matched HCs were recruited from surrounding communities. Meanwhile, all patients met the criteria for MDD based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) [19], as confirmed by two experienced psychiatrists using the Structured Clinical Interview. Exclusion criteria for CSD patients included: (i) age below 18 years or left-handedness; (ii) prior history of neoplasms or other intracranial diseases; (iii) known psychiatric symptoms preceding COVID-19 infection; (iv) organic diseases or severe substance dependence (e.g., tobacco, drugs, or alcohol); and (v) contraindications for MRI scanning (e.g., claustrophobia or metallic implants). The HCs were screened without any personal or familial history of psychiatric disorders, no evidence of organic brain disease, and no prior use of psychiatric medications. To exclude asymptomatic COVID-19 infections, individuals with positive nucleic acid PCR detection were further removed from the HCs cohorts. Demographic data, such as age, gender, and educational levels, were recorded for all participants. Educational attainment was categorized into four levels: below senior high school, senior high school, undergraduate, and above undergraduate. The protocol of this study was approved by the Ethics Committee of The Affiliated Yueqing Hospital of Wenzhou Medical University (ID: YQYY202500055), and written informed consents were acquired from all participants prior to the extensive screening procedure.

Psychological examination

Comprehensive psychological evaluations were performed on each CSD patient to explore possible links between neuroimaging features and clinical condition. Symptomatology was assessed using validated electronic questionnaires administered via the Wen Juan Xing platform (Ranxing Information Technology Co., Ltd., Changsha, China). The instruments included the Patient Health Questionnaire-9 (PHQ-9) for depression [20], the Generalized Anxiety Disorder-7 (GAD-7) scale for anxiety [21], the Perceived Stress Scale-14 (PSS-14) for psychological stress symptoms [22], and the Athens Insomnia Scale (AIS) for sleep disturbances [23]. All assessments were performed under the supervision of an experienced psychiatrist to ensure the accuracy, reliability, and consistency of the data collection process.

Neuroimaging acquisition and preprocessing

The 3.0 straticulate Siemens MRI scanner was applied to obtain brain images from all participants in the Affiliated Yueqing Hospital of Wenzhou Medical University. High-resolution three-dimensional T1-weighted images (3D-T1WI) were obtained in sagittal plane using a 3D-BRAVO sequence according to following parameters: repetition time (TR) = 8 ms, echo time (TE) = 3.84 ms, flip angle = 12°, acquisition matrix = 256 × 256, field of view (FOV) = 256 × 256 mm², number of slices = 196, and slice thickness = 1 mm. Subsequently, the 3D-T1WI data were preprocessed using the recon-all pipeline from FreeSurfer software version 6.0 (http://surfer.nmr.mgh.harvard.edu/) within a surface-based framework [24]. Structural preprocessing included skull removal, tissue’s segmentation, segmentation of hemispheric and subcortical signatures, and reconstruction of gray-white matter boundaries and cortical surfaces.

MIND network construction

The MIND framework was designed to quantify inter-regional similarity of cortical morphology across individuals. Cortical regions were initially defined using the 68-region Desikan-Killiany (DK) atlas. To achieve finer spatial granularity, each region was further subdivided into spatially contiguous subunits via a backtracking algorithm [25]. This subdivision yielded the DK-308 parcellation, which was subsequently projected onto each participant’s reconstructed cortical surface. For each of the 308 parcels, five morphometric features were extracted to capture complementary aspects of cortical architecture: gray matter volume, mean curvature, surface area, cortical thickness, and sulcal depth [26]. Subsequently, these features were standardized within each participant using z-score normalization, producing a multivariate morphometric profile for every parcel. Finally, the multivariate distribution of features for each parcel was compared pairwise using the Kullback-Leibler (K-L) divergence. Given the asymmetry of K-L divergence, MIND defined similarity by computing the inverse of the divergence, thereby emphasizing regions with comparable morphometric distributions while reducing the impact of outliers [15]. This procedure generated an individual-specific 308 × 308MIND similarity matrix for each participant, which served as the basis for subsequent network analyses. A schematic overview of the MIND construction workflow is presented in Supplementary Fig. 1, and the 308 brain regions/parcellation schemes were listed in Supplementary Table. 1. The implementation code is publicly available at https://github.com/isebenius/MIND.

Case-control regional mind difference

To assess regional MIND differences between CSD individuals and HCs, we applied a general linear model (GLM) incorporating age, sex, age×sex and total intracranial volume (TIV) as covariates. Two-sample t-test (contrast: CSD vs HCs) was then conducted, and significant MIND alterations were further contextualized by mapping all 308 cortical regions onto two established neuroanatomical frameworks: the Yeo-7 functional network [27] and the von Economo cytoarchitectonic atlas [28]. To investigate the relationship between regional MIND alterations and disease severity, CSD patients were stratified into different subgroups based on psychological scale scores and infection times before the diagnosis of depression. Subsequently, the spatial correlations between case-control t-values and regional MIND values were calculated to assess spatial specificity of alterations. All statistical comparisons were conducted within the aforementioned GLM framework, and a false discovery rate (FDR)-corrected p-value < 0.05 was considered statistically significant for multiple comparisons.

Cognitive terms associated with regional mind alterations

The Neurosynth Python tool (https://neurosynth.org/) was employed to investigate potential associations between regional MIND alterations and meta-analytic cognitive terms in CSD patients [29]. Following the removal of non-cognitive terms, 123 terms related to cognition were kept from the Neurosynth database, and thresholded t-maps from MIND group comparisons were divided into two categories: CSD-positive and CSD-negative. The “decoder” function was further applied to compute the spatial correlations between these t-maps and the meta-analytic maps with each cognitive term. The top 30 cognitive terms with the highest correlation coefficients using spin tests were selected for further interpretation.

Transcriptomic profiles acquisition and preprocessing

Transcriptomic profiles from 3,702 brain regional samples, derived from six postmortem donors, were obtained from the Allen Human Brain Atlas (AHBA) database (http://human.brain-map.org). The average age of the donors was 42.50 ± 13.38 years, and the postmortem brain samples consisted of two from the right hemisphere and six from the left hemisphere. To ensure uniformity, only left hemisphere tissues were included in the subsequent analysis. The “abagen” Python toolbox [30] was employed to map the transcriptomic profiles onto 200 corresponding regions within the left hemisphere, and the gene expression datasets were further normalized using a standardized seven-step preprocessing protocol. This process yielded a final transcriptional matrix comprising 200 brain regions and 15,631 genes.

Transcription-regional mind spatial correlation analysis

To examine the spatial relationship between transcriptomic expression and regional MIND characteristics in CSD patients, partial least squares (PLS) regression utilized the expression of 15,631 genes as independent variables and the case-control t-maps of 200 regional MIND values as dependent variables [31]. The fourth PLS component (PLS4) was selected as the optimal low-dimensional representation of covariance within the high-dimensional data matrix. Subsequently, spatial autocorrelation was assessed through a 10,000-iteration permutation test to determine whether the observed covariance between the MIND t-statistic maps and PLS4 weights exceeded that expected by chance. Moreover, gene variability in PLS4 was evaluated using 10,000 bootstrap resamples, with Z-scores calculated by dividing the regional weights by their corresponding bootstrap standard errors [32]. Genes were subsequently ranked, and those with FDR-corrected p < 0.05 were identified as significantly PLS4-weighted. Finally, Spearman correlation was applied to examine spatial associations between regional MIND t-statistics and PLS4 weights.

Functional annotation and potential association with other psychological disorders

To elucidate the functional characteristics of PLS-weighted genes, we performed the functional enrichment analysis using Metascape v3.5 (https://metascape.org/), incorporating annotations from the Gene Oncology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and DisGeNET databases [33]. Significantly enriched pathways were defined as FDR-adjusted p-value < 0.05 and enriched gene counts > 5. Moreover, to investigate potential associations between aberrant MIND patterns in CSD patients and gene dysregulation in other psychiatric conditions, we compared PLS4-weighted genes with DEGs from six major disorders reported by Gandal et al. [34], including autism spectrum disorder (ASD), adult-SCZ, major depressive disorder (MDD), alcohol abuse disorder (AAD), bipolar disorder (BD), and inflammatory bowel disease (IBD). Disorder-specific DEGs were defined by log2(Fold change) > 0 and FDR-adjusted p-value < 0.05. The significance of overlap between PLS4 genes and each disorder’s DEGs was evaluated via permutation testing (10,000 iterations), assessing whether the mean PLS4 Z-score of shared genes exceeded random expectations. A one-sided permutation p-value with FDR correction < 0.05 was considered statistically significant.

Protein-protein interaction (ppi) network analysis for PLS weighted genes

To validate the functional characteristics at protein aspects, we conducted the PPI network analysis using STRING online-tool (https://string-db.org/) based on the PLS-weighted genes. The Cytoscape software was applied to construct the PPI networks and the “NetworkAnalyzer” function was employed to screen key nodes with significant degree and co-expression values. Finally, the GO enrichment analysis was utilized to validate the functional enrichments of key proteins.

Assigning CSD-related transcriptional patterns to special cortical layers, cell subtypes, and developmental phases

To assign CSD-related (PLS4-weighted) genes to specific cortical layers, cell subtypes, and developmental stages, we conducted a comprehensive multi-dimensional enrichment analysis. Gene Set Enrichment Analysis (GSEA), implemented via the “GSEABase” and “clusterProfiler” packages [35], was employed to examine associations between MIND alterations and six cortical layers, based on transcriptomic signatures from a published transcriptomic reference. Additionally, to evaluate temporal specificity, the cell-type specific expression analysis (CSEA) tool [36] was employed to assess developmental stage enrichment across brain regions. Cell-type-specific enrichment was assessed using marker gene lists from five independent single-cell RNA-sequencing studies, covering microglia, oligodendrocyte precursor cells (OPCs), endothelial cells, oligodendrocytes, astrocytes, and excitatory/inhibitory neurons. Single-sample GSEA (ssGSEA) was then performed using the GSVA package with a Gaussian kernel to quantify cellular composition in each brain region [37]. To confirm cell-type specificity associated with regional MIND variation, we overlapped PLS4-weighted genes with cell-type markers and validated the results through permutation testing. Functional enrichment of overlapped gene sets within each cell type was further examined using the Metascape platform.

Null model and reproducibility validation

To address potential confounding effects from spatial autocorrelation, we applied a spin test-based null model, which generates null Spearman correlation coefficients by rotating cortical maps on a spherical surface while preserving spatial topology [38]. A null distribution was constructed using 10,000 spin permutations across cortical parcels. The resulting P_spin value was defined as the proportion of permuted correlations exceeding the observed correlation coefficient. To ensure robustness, we performed several validation analyses: (i) assessing the stability of case-control MIND t-maps without adjusting for TIV; (ii) evaluating the effect of varying connection density thresholds in the MIND matrix (ranging from 10% to 90% in 20% increments); and (iii) validating functional enrichment reliability through a multi-gene list meta-analysis incorporating depression-associated genes from genome-wide association studies (GWAS), including anxiety disorder (AD), MDD, and insomnia [39]. These procedures collectively enhance the confidence in the spatial specificity and biological relevance of our findings.

Results

Demographic and clinical characteristics

This study investigated MIND alterations in CSD patients, examining their associations with cognitive domains, transcriptional patterns, specific cell-subtypes, cortical layers, and developmental phases (Fig. 1). Demographic comparisons between CSD patients (average age: 49.80 ± 13.11 years) and HCs (average age: 47.35 ± 14.44 years) revealed no statistically significant differences in age or sex. In addition, no significant difference was observed in educational history between the CSD and HC cohorts (Chi-square test, p = 0.876), thereby minimizing the potential influence of intellectual ability (Table. 1). The CSD patients were stratified into subgroups based on infection history: 42 with single COVID-19 infection and 38 with reinfection, as well as according to disease severity measured by various psychological scales. Moreover, further comparisons between the once-infected and reinfected groups showed no significant differences in age, sex, educational background, or AIS insomnia scores (Chi-square test, p > 0.05). However, patients with reinfection exhibited a significantly higher proportion of severe symptoms, including elevated scores on the PSS-14 (23.68% vs 0.00%), PHQ-9 (100% vs 0.00%) and GAD-7 scale (65.79% vs 0.00%) scales (Table. 2).

Fig. 1.

The workflow of the study design. (a) construction of mind networks. Multiple macrostructural features (gray matter (gm), surface area (sa), sulcal depth (sd), mean curvature (MC), and cortical thickness (ct)) constructed the 308 × 308matrix for mind. Then, the regional mind values were established by averaging all the connections of the 308 cortical areas without applying any thresholding. (b) case-control difference of regional mind alterations and associations with functional networks and distinct clinical grades. (c) transcriptional signature assessment for genes related to alterations in mind. PLS regression was then used to identify imaging transcriptomic associations, including similarity evaluation with other psychiatric disorders, functional enrichment, ppi network analysis, meta-analytic cognitive terms, specific cell-subtypes, cortical layer distribution, and stage-dependent developmental susceptibility. mind, morphometric inverse Divergence; PLS, partial least squares; ppi, protein-protein interaction

Table 1.

The clinical characteristics of subject population

Characteristics	CSD (n=80)	HC (n=40)	P value
Age (year), mean ± sd	49.80±13.11	47.35±14.44	0.550
Gender			0.950
female, n (%)	28 (35.00%)	14 (35.00%)
male, n (%)	52 (65.00%)	26 (65.00%)
Education
junior school, n (%)	3 (3.75%)	2 (5.00%)	0.876
senior school, n (%)	50 (62.50%)	24 (60.00%)
undergraduate, n (%)	22 (27.50%)	10 (25.00)
postgraduate, n (%)	5 (6.25%)	4 (10.00%)
Infection condition
once infection, n (%)	42 (52.50%)
twice infection, n (%)	38 (47.50%)
PSS-14 stress score
normal (0-28), n (%)	28 (35.00%)
mild (29-42), n (%)	42 (52.50%)
severe (43-56), n (%)	10 (12.50%)
PHQ-9 depression score
normal (0-4), n (%)	0 (0.00%)
mild (5-9), n (%)	28 (35.00%)
middle (10-14), n (%)	14 (17.50%)
severe (15-27), n (%)	38 (47.50%)
GAD-7 anxiety score
normal (0-4), n (%)	0 (0.00%)
mild (5-9), n (%)	36 (45.00%)
middle (10-14), n (%)	19 (23.75%)
severe (15-21), n (%)	25 (31.25%)
AIS insomnia score
normal (0-3), n (%)	0 (0.00%)
suspicious (4-6), n (%)	6 (7.50%)
insomnia (>6), n (%)	74 (92.50%)

Note: CSD, COVID-19 secondary depression; HC, health control; PSS-14, perceived stress 14-item scale; PHQ-9, patient health 9-item questionnaire; GAD-7, general anxiety disorder 7-item scale; AIS, athens insomnia scale

Table 2.

Comparison of clinical characteristics between CSD patients with single and twice infection

Characteristics	CSD-single (n =42)	CSD-twice (n =38)	P value
Age (year), mean ± sd	51.22±14.03	48.21±11.16	0.340
Gender			0.050
female, n (%)	9 (22.22%)	19 (50.00%)
male, n (%)	33 (77.78%)	19 (50.00%)
Education			0.699
junior school, n (%)	2 (4.76%)	1 (2.63%)
senior school, n (%)	23 (54.76%)	27 (71.05%)
undergraduate, n (%)	14 (33.33%)	8 (21.05%)
postgraduate, n (%)	3 (7.14%)	2 (5.26%)
PSS-14 stress score			<0.001***
normal (0-28), n (%)	28 (66.67%)	0 (0.00%)
mild (29-42), n (%)	14 (33.33%)	29 (76.32%)
severe (43-56), n (%)	0 (0.00%)	9 (23.68%)
PHQ-9 depression score			<0.001***
normal (0-4), n (%)	0 (0.00%)	0 (0.00%)
mild (5-9), n (%)	28 (66.67%)	0 (0.00%)
middle (10-14), n (%)	14 (33.33%)	0 (0.00%)
severe (15-27), n (%)	0 (0.00%)	38 (100%)
GAD-7 anxiety score			<0.001***
normal (0-4), n (%)	0 (0.00%)	0 (0.00%)
mild (5-9), n (%)	33 (78.57%)	3 (7.89%)
middle (10-14), n (%)	9 (21.43%)	10 (26.32%)
severe (15-21), n (%)	0 (0.00%)	25 (65.79%)
AIS insomnia score			0.050
normal (0-3), n (%)	0 (0.00%)	0 (0.00%)
suspicious (4-6), n (%)	6 (14.29%)	0 (0.00%)
insomnia (>6), n (%)	36 (85.71%)	38 (100%)

Note: CSD, COVID-19 secondary depression; HC, health control; PSS-14, perceived stress 14-item scale; PHQ-9, patient health 9-item questionnaire; GAD-7, general anxiety disorder 7-item scale; AIS, athens insomnia scale

Regional mind differences in CSD

The global MIND was computed as the mean of the regional MIND values across all cortical regions. After controlling for age, sex, and TIV, a significant difference in the global MIND distribution was observed between CSD patients and HCs, as determined by a two-sample Kolmogorov-Smirnov test (p < 0.001, Fig. 2a). The spatial distribution of regional MIND in CSD patients resembled that of HCs, with elevated values in the temporoparietal and frontal cortices and reduced values in the insular cortex (Fig. 2b). Regionally, CSD patients possessed significantly increased MIND in the left caudal anterior cingulate (part1, t = 3.77, p = 0.032), left isthmus cingulate (part1, t = 5.46, p < 0.001), left middle temporal cortices (part5, t = 3.61, p = 0.032), left posterior cingulate (part2, t = 3.73, p = 0.032), right isthmus cingulate (part2, t = 3.57, p = 0.032) and right supramarginal regions (part4, t = 3.69, p = 0.032) (Fig. 2c, Supplementary Table 2). Interestingly, a positive spatial correlation was found between mean control MIND values and CSD-HCs t-maps (r₍₃₀₈₎ = 0.11, P_spin = 0.031), indicating that regions with higher normative connectivity showed greater case-control alterations (Fig. 2d). Additionally, t-values from the comparison between CSD and HCs were strongly correlated with those from single-infection (r₍₃₀₈₎ = 0.89, P_spin < 0.001) and reinfection subgroups (r₍₃₀₈₎ = 0.83, P_spin < 0.001) (Supplementary Fig. 2a-b).

Fig. 2.

Case-control regional mind alterations for CSD patients. (a) the density diagram displays the frequency distributions of mean mind values between the CSD individuals and HCs after removing the influence of age, sex, age ×sex, and tiv. (b) the mean mind maps of 308 brain regions in CSD patients and HCs. The spatial distribution of regional mind in CSD patients resembled that of HCs, with elevated values in the temporoparietal and frontal cortices and reduced values in the insular cortex. (c) case-control t-maps for CSD patients vs HCs. CSD patients exhibited significantly increased mind in the left caudal anterior cingulate (part1), left isthmus cingulate (part1), left middle temporal cortices (part5), left posterior cingulate (part2), right isthmus cingulate (part2) and right supramarginal regions (part4). (d) CSD-HCs t-maps showed a positive spatial correlation with mean control mind values (r₍₃₀₈₎ = 0.11, P_spin = 0.031). (e) comparison of critical mind regions among different CSD subgroups within distinct clinical grades. P_spin, spin test

Furthermore, compared to mild or moderate CSD patients, those classified as severe exhibited increased MIND values in the left isthmus cingulate (part1), while decreased MIND values in the right isthmus cingulate (part2) (Fig. 2e). However, no statistically significant differences were observed in other regions among severity-based subgroups as defined by clinical scale scores (Supplementary Fig. 3a-d). Additionally, we identified aberrant MIND patterns within both functional Yeo-7 networks and von Economo frameworks in CSD patients as compared to HCs. Within the Yeo-7 network parcellation, CSD patients demonstrated significantly elevated MIND in six functional networks except the visual network, including dorsal attention, default mode, frontoparietal, limbic, somatomotor, and ventral attention networks (Fig. 3a). In terms of the von Economo cortical classification, significantly increased MIND was observed in association cortex I/II, insula cortex, limbic cortex, and primary motor cortex, with the exception of the primary or secondary sensory cortices (Fig. 3b). Notably, both single-infection and reinfection subgroups exhibited elevated MIND alterations in above key cortical regions (Supplementary Fig. 3e) and functional/cytoarchitectonic networks (Supplementary Fig. 4a-b) compared to HCs, while there was no significant difference between single-infection and reinfection, suggesting that regional MIND changes are not dependent on the times of COVID-19 infections.

Fig. 3.

Functional networks, meta-analytic cognitive terms, and transcriptional signatures related to CSD-HCs regional MIND alterations. (a-b) Comparison of both functional Yeo-7 networks and von Economo frameworks in CSD patients as compared to HCs. (c-d) Word-cloud diagram showed cognitive terms associated with CSD-HCs regional MIND t-maps. (e) The consistent distributions between CSD-HCs regional MIND t-maps and PLS4 score maps. (f) PLS4 scores exhibited a significant positive spatial correlation with regional t-values (r =0.36, P_spin <0.001), indicating that genes with high PLS4 weights are predominantly expressed in areas exhibiting elevated MIND in CSD. (g) The univariate Z-test identified a total of 2,296 PLS2 weighted genes with FDR-corrected P <0.05, including 1152 PLS4+ genes (Z >2.68) and 1144 PLS4- genes (Z <-2.68). (h) Combined with dysregulated gene sets from six major psychiatric conditions, significant enrichment of PLS4-weighted genes was found in BD, IBD and MDD, but not in ASD, AAD and SCZ (FDR-corrected P_perm <0.05). P_spin, spin test; P_perm, permutation test. ASD, autism spectrum disorder; SCZ, schizophrenia; MDD, major depressive disorder; AAD, alcohol abuse disorder; BD, bipolar disorder; IBD, inflammatory bowel disease.

Meta-analytic cognitive terms associated with mind abnormalities

Our investigation examined the associations between CSD-HCs t-maps of regional MIND and meta-analytic cognitive terms. The t-map, reflecting increased MIND in CSD patients, was predominantly correlated with cognitive functions related to auditory processing, semantic interpretation, empathy, semantic memory, familiarity, and language comprehension (Fig. 3c). In contrast, the t-map indicating regionally decreased MIND in CSD patients was primarily linked to visual-cognitive domains, including face perception, spatial localization, object recognition, and gaze processing (Fig. 3d).

Transcriptional features related to regional mind changes

Using the transcriptional matrix encompassing 152 brain regions and 15,631 genes, we conducted PLS regression to unveil transcriptional patterns associated with the spatial distribution of MIND CSD-HCs t-maps. The PLS4 explained 17.68% of the variance in macrostructural alterations observed in CSD patients, a proportion significantly higher than expected by chance (P_spin < 0.001, Supplementary Fig. 5). Notably, the spatial distribution of PLS4 scores closely mirrored the MIND t-maps (Fig. 3e) and exhibited a significant positive spatial correlation with regional t-values (r = 0.36, P_spin < 0.001), indicating that genes with high PLS4 weights are predominantly expressed in areas exhibiting elevated MIND in CSD (Fig. 3f). The univariate Z-test identified a total of 2,296 PLS2 weighted genes with FDR-corrected p < 0.05, including 1152 PLS4+ genes (Z > 2.68) and 1144 PLS4- genes (Z < −2.68) (Fig. 3g). Combined with dysregulated gene sets from six major psychiatric conditions, significant enrichment of PLS4-weighted genes was found in BD, IBD and MDD, but not in ASD, AAD and SCZ (FDR-corrected P_perm < 0.05) (Fig. 3h). These findings underscore potential molecular associations between CSD-related regional MIND alterations and genetic dysregulation in BD, IBD, and MDD.

Functional enrichment of PLS4 weighted genes associated with regional mind alterations

To further elucidate transcriptional patterns linked to CSD-related regional MIND alterations, functional enrichment analyses were conducted using Metascape based on PLS4± gene-lists. The PLS4+ genes were significantly enriched in neurodevelopmental and metabolism-related pathways, including “Naba Core Matrisome”, “neuron projection development”, “organophosphate biosynthetic process”, “head development”, “phospholipid metabolic process” (Fig. 4a–b). In contrast, PLS4- genes showed prominent enrichment in immune response-related pathways, including “positive regulation of immune response”, “innate immune response”, “Adaptive Immune System”, and “Cytokine Signaling in Immune system” (Fig. 4c–d). To validate these functional profiles at protein’s level, we respectively constructed two comprehensive PPI networks and screened key hub proteins with high degree and co-expression levels. Notably, core proteins such as HDAC3, ATP5PO, ABL1, and RBM39 were associated with “phosphorus metabolic process” and “neuron development” in the PLS4+ group, while MYC, PTPRC, TLR4, and ITGAM were enriched in immune-related pathways for the PLS4- group Supplementary Fig. 6a-b). These results suggest that disruptions in neurodevelopmental metabolism and immune regulation may collectively contribute to regional MIND alterations observed in CSD patients.

Fig. 4.

Functional annotation of PLS4-weighted signatures. (a/c). the bar diagram displayed the enrichment of go terms and kegg pathways, including neurodevelopmental and metabolism-related pathways for PLS4+ genes (a) and immune response-related pathways for PLS4- genes (c) (b/d). the enrichment network comprised of inter-nodes and intra-nodes similarities of enriched processes and pathways for PLS4+ (b) and PLS4- weighted genes (d) using the metascape software. The nodes and lines represent terms/pathways and links among terms, respectively, and the size of nodes represents the number of enriched genes. go, gene Oncology; kegg, Kyoto encyclopedia of genes and genomes

Transcriptional enrichment for specific cortical layers, cell subtypes, and developmental stages

To further elucidate the transcriptional signatures associated with specific cortical architectures and MIND changes in CSD, we utilized PLS4-weighted genes to assess layer-specific enrichment across six cortical layers using GSEA algorithm. The PLS4+ genes demonstrated significant enrichment in layer-V (NES = 1.89, adjusted-p = 0.014, Supplementary Fig. 7a), whereas the PLS4- genes were primarily enriched in layer-I (NES = −2.18, p = 0.001, Fig. 5a). Developmental gene expression patterns were further examined using the CSEA tool, which revealed that PLS4- genes were predominantly expressed during the early fetal (EF) stages, especially in specific regions such as striatum, hippocampus, cortex, and amygdala (Fig. 5b). In contrast, PLS4+ genes were enriched during late infancy (LI) to young adulthood (YA), particularly within the hippocampus, cortex and cerebellum (Supplementary Fig. 7b). Using the BrainSpan dataset, we extracted a transcriptional matrix comprising 12,676 genes across 12 brain regions and five developmental stages from fetal to adult, and further calculated regional gene scores as developmental stage-specific signatures. Notably, these gene scores exhibited age-dependent variation, with notable transitions between early and late stages, except during the child and adolescent periods (Fig. 5c). Significant spatial correlations were observed between AHBA-derived and BrainSpan-defined gene scores within the fetal (r = 0.52, P_perm = 0.047), infant (r = 0.56, P_perm = 0.032), and adult (r = 0.59, P_perm = 0.021) stages, confirming transcriptional consistency across different datasets.

Fig. 5.

Transcriptional signatures evaluation for specific cell-subtypes, cortical layer distribution, and stage-dependent developmental susceptibility using PLS4- weighted genes. (a) PLS4- genes demonstrated significant enrichment in layer-I (nes = −2.18, p = 0.001). (b) the CSEA analysis revealed that PLS4- genes were predominantly expressed during the early fetal (ef) stages, especially in specific regions such as striatum, hippocampus, cortex, and amygdala. (c) based on BrainSpain database, regional gene scores exhibited age-dependent variation, with notable transitions between early and late stages, except during the child and adolescent periods. Significant spatial correlations were observed between AHBA-derived and BrainSpan-defined gene scores within the fetal (r = 0.52, P_perm = 0.047), infant (r = 0.56, P_perm = 0.032), and adult (r = 0.59, P_perm = 0.021) stages. (d) ssGSEA analysis revealed significant associations between PLS4- genes and both microglia (number = 23, FDR-corrected P_perm < 0.001) and astrocytes (number = 23, FDR-corrected P_perm = 0.021). (e-f) functional enrichments and interactional networks for overlapped genes for different cell subtypes. The color of nodes represents distinct cell-types, and the size of nodes represents the number of enriched genes. nes, normalized enrichment score; CSEA, cell-type specific expression analysis; ssGSEA, single-sample gene set enrichment Analysis; P_perm, permutation test

To explore cell-type-specific transcriptional patterns associated with MIND alterations in CSD, ssGSEA analysis revealed significant associations between PLS4- genes and both microglia (number = 23, FDR-corrected P_perm < 0.001) and astrocytes (number = 23, FDR-corrected P_perm = 0.021) (Fig. 5d), with no significant associations observed for PLS4+ genes (Supplementary Fig. 7c). Notably, the enrichment analysis further demonstrated both microglia and astrocytes contribute to cellular migration-related processes, including “positive regulation of programmed cell death”, “cell-cell adhesion”, and “positive regulation of cell migration”. Additionally, microglia were specifically implicated in immune response-related pathways, such as “Regulatory circuits of the STAT3 signaling pathway”, “Interleukin-2 family signaling”, and “positive regulation of immune response” (Fig. 5e–f).

Robustness validation of CSD-related mind alterations and transcriptional signatures

To ensure the robustness of our results, we assessed whether TIV influenced the observed case-control differences in global MIND. The MIND disparities between patients and controls adjusted for TIV were highly covariate, showed a high consistent with those calculated without TIV correction (r = 0.99, P_spin < 0.001; Supplementary Fig. 8), indicating minimal confounding from TIV. We further calculated MIND matrices across a spectrum of connection densities, confirming that both the absolute MIND maps and their corresponding t-value distributions remained stable and comparable to those generated at full (100%) connectivity (Supplementary Fig. 9–10). To further validate the biological relevance of our enriched pathways, we conducted a multi-angle meta-analysis based on the GWAS gene sets related to AD, MDD, and insomnia. This analysis identified several overlapping neurodevelopmental pathways shared between depression-related GWAS loci and PLS4± weighted genes, encompassing terms like “regulation of cell projection organization”, “neuron projection development”, “head development”, and “regulation of synapse structure or activity”. Additionally, immune-related pathways were predominantly associated with PLS4- genes, while both PLS4+ and PLS4- genes showed significant enrichment in metabolic pathways, including “positive regulation of phosphorus metabolic process” and “carboxylic acid metabolic process” (Supplementary Fig. 11a-b). These enrichment patterns were largely concordant with those observed in the functional analyses of PLS4 weighted genes. Collectively, these results support the genetic validity and biological plausibility of our findings, suggesting that the PLS-weighted genes are functionally linked to disease-relevant GWAS loci and may underlie region-specific alterations in MIND in CSD patients (Supplementary Fig. 11c-d).

Discussion

This study provides a comprehensive analysis of the interplay between regional MIND abnormalities and potential transcriptomic mechanisms in CSD patients. Our cohort consisted of 80 first-episode, treatment-naïve CSD patients and 40 age- and sex-matched HCs. Our findings demonstrate that transcriptional similarity architecture, together with structural connectivity, constrains the spatial patterning of MIND alterations in CSD. These alterations were further contextualized by their associations with clinical scale scores and cognitive terms. Moreover, we identified significant spatial correlations between the MIND case-control t-map and cortical gene expression patterns represented by the PLS4 weights. Additionally, A substantial overlap was observed between PLS4-weighted genes and dysregulated genes previously implicated in MDD, BD, and IBD. Genes linked to MIND abnormalities exhibited a significant enrichment not only in neurodevelopmental and metabolic processes (PLS4+ genes), but also in immune response-related pathways (PLS4- genes). Furthermore, advanced enrichment analysis further indicated that PLS4- genes showed preferential expression in cortical layer-I, specific microglia and astrocytes, along with susceptibility during EF stages. Conversely, PLS4+ genes exhibited significant enrichment in cortical layer-V, and susceptibility during LI to YA stages, without notable cell-type specificity. These findings highlight the multiscale pathophysiological architecture of CSD, linking macroscale morphometric alterations to molecular and cellular transcriptional signatures.

Unlike traditional morphometric approaches based on isolated morphometric metrics or regional summary statistics, MIND quantifies structural similarity by capturing divergence between multidimensional distributions, offering heightened sensitivity to individual variability arising from developmental and genetic factors. In this study, CSD patients demonstrated significantly elevated regional MIND values, particularly within the left anterior and isthmus cingulate (involved in emotional and cognitive regulation), left posterior cingulate (linked to memory and visuospatial function), left middle temporal cortices (related to auditory and language processing), and right supramarginal regions (engaged in language comprehension). These increased MIND values may reflect diminished structural differentiation in CSD patients, potentially resulting in functional dysregulation in corresponding regions [40, 41]. Our meta-analysis confirmed that brain regions exhibiting prominent MIND alterations in CSD patients were significantly linked to various high-order cognitive functions, including listening, meaning, semantic memory, face recognition, spatial localization, and object recognition, consistent with prior findings [42]. These results underscore a potential link between impaired regional MIND and cognitive deficits in individuals with CSD. Moreover, widespread MIND abnormalities observed across both functional and cellular networks further suggest early-stage, multisystem network dysregulation in CSD. Notably, regional MIND alterations were positively correlated with perceived stress scores and negatively correlated with anxiety, depression, and insomnia scores, underscoring their relevance to clinical severity. Conversely, no significant associations were identified between abnormal MIND alterations and infection frequency, indicating that reinfection couldn’t exacerbate macroscale morphometric disruptions in CSD patients. Additionally, reproducibility analyses confirmed that the MIND differences remained stable regardless of TIV or connection density, supporting the robustness and generalizability of our findings.

Previous GWAS studies have highlighted shared genetic architectures among major psychiatric disorders, particularly in ASD, BD, MDD, and schizophrenia [43]. Dirk et al. further demonstrated substantial overlap in differentially expressed genes across these conditions, reinforcing the shared genetic risks among psychiatric disorders [44]. Beyond MDD and BD, our genetic convergence analysis revealed that PLS4-weighted genes were significantly associated with dysregulated genes from IBD, suggesting shared genetic commonality and potentially overlapping pathophysiological mechanisms among these psychiatric conditions. Notably, a systematic review reported depression prevalence rates of 34.7% in active IBD and 19.9% in non-active IBD patients [45]. Moreover, transcriptomic profiling also identified key diagnostic genes (SPARC, HGF, MMP8, and ADAM12) with high diagnostic efficiency for IBD-related major depressive disorder [46]. Collectively, these findings support the hypothesis that diverse psychiatric conditions may converge on common neurobiological substrates, potentially mediated through shared immune pathways or the brain-gut axis.

Functional annotation of PLS4-weighted signatures provided critical insights into the transcriptional patterns linked to regional MIND alterations in CSD patients. Intriguingly, PLS4+ genes exhibited a significantly enrichment in neurodevelopmental and metabolism-related functions, including “neuron projection development”, “head development”, and “phospholipid metabolic process”. Neuronal development is fundamental for synaptic maturation and stability, and has been implicated in the pathogenesis of depression through disruptions in hippocampal, putaminal, and amygdalar circuits during early to mid-adolescence [47]. Phospholipid metabolism, essential for maintaining neuronal and synaptic integrity, also plays a critical role in neurotransmitter signaling, particularly involving dopamine, serotonin, glutamate, and acetylcholine. The dysregulation of phospholipid metabolism has been implicated in the complex etiology of MDD [48]. Contrastively, PLS4- genes were predominantly enriched in immune response-related pathways, indicating that potential activation of inflammation and immune responses may underlie aspects of CSD pathophysiology [49]. Furthermore, PPI network analysis further identified critical modules centered around phosphorus metabolism, neuronal development, and immune regulation. Additionally, multi-gene-list meta-analysis incorporating GWAS-derived MDD risk genes validated the robustness of these enrichments, particularly within neurobiological categories corresponding to PLS4+ gene functions. Overall, these results underscore the crucial role of genetic influences in shaping large-scale neural architecture in CSD, and highlight that disruptions in both neurodevelopmental and immune pathways may serve as fundamental mechanisms driving disease onset and progression.

Cellular dysfunction and the spatial organization of cortical layers are critically implicated in the pathophysiology of various psychiatric conditions [50]. Utilizing cell type- and cortical layer-specific transcriptional markers, we detected significant enrichments of PLS4- genes in microglia and astrocytes, whereas PLS4+ genes showed no such association. Emerging evidences have emphasized microglia as critical mediators of depressive pathology, primarily through regulating neuroinflammation, synaptic remodeling, and neural circuit formation [51]. Similarly, astrocytes essential for neurodevelopment and homeostatic regulation, contribute to depression by modulating neuronal number, morphology, and purinergic signaling pathways [52]. Additionally, our layer-specific enrichment analysis revealed that PLS4+ genes were predominantly enriched in cortical layer-V, while PLS4- genes showed significant enrichment in cortical layer-I. These laminar-specific distributions may reflect neuronal subtype- and region-specific vulnerabilities contributing to CSD pathology [53]. Complementary analyses using CSEA and BrainSpan datasets revealed that aberrant expression of PLS4-weighted genes prominently disrupted within early (fetal, infant) and posterior (adult) developmental stages, rather than childhood or adolescent stages, suggesting temporally distinct windows of susceptibility driven by transcriptional dysregulation aligned with region-specific MIND alterations [54]. These findings underscore the importance of microscale cortical disruptions in the neurobiological underpinnings of CSD, encompassing aberrations in glial cell function, cortical layer-specific gene expression, and stage-dependent developmental susceptibility.

Several limitations within our study warrant consideration. First, the sample size of our cohorts was correspondingly modest, and larger-scale studies are warranted to further elucidate and validate the associations between MIND alterations and specific transcriptional signatures in CSD. Second, transcriptional profiles were derived from the AHBA database, which included only two right-hemisphere samples, potentially introducing hemispheric biases. Although subsequent analyses were restricted to left hemisphere data, the relationship between transcriptional signatures and MIND alterations in the right hemisphere remains unexplored. Additionally, due to practical limitations, we were unable to obtain sufficient brain imaging data from pure-depression patients. Consequently, direct comparisons of MIND differences between CSD and depression patients could not be performed, and the current findings are limited to differences observed between CSD patients and HC cohorts. Moreover, although the MIND framework provides greater stability, improved alignment with cortical cytoarchitecture, and increased heritability compared to traditional morphological studies, it remains limited in evaluating cortical structural changes, lacking the assessment for subcortical structures. Finally, while we conducted integrative analyses linking MIND abnormities to specific cell subtypes, cortical layers, and developmental phases, the specific regulatory mechanisms remain unclear. Further experimental validation using in vivo and in vitro models is necessary to uncover the molecular pathways underlying MIND abnormalities in CSD.

Conclusions

In summary, this study demonstrated that regional MIND alterations in individuals with CSD are spatially linked to specific transcriptional signatures. PLS-weighted genes were significant enriched in pathways related to neurodevelopment, phospholipid metabolism and immune response, highlighting shared molecular mechanisms with other psychiatric disorders (MDD, BD, and IBD). Notably, these transcriptomic disturbances were associated with macroscale morphological alterations in MIND, particularly involving microglia, astrocytes, and cortical layers-I/V, as well as stage-specific developmental susceptibility. Collectively, our findings provide novel insights into the multiscale pathological architecture of CSD, bridging cortical morphometric disruptions with underlying gene expression profiles. This integrated framework enhances our understanding of the molecular basis of brain morphometric disruptions in CSD and may inform future research into targeted therapeutic strategies.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Chenlu Li: Data curation, Formal analysis, Methodology, Visualization, Writing-original draft, and Writing-review & editing. Qingshi Lin: Data curation, Methodology, and Writing-original draft. Liaoliao Yang: Conceptualization, Investigation, Formal analysis, Funding acquisition, Writing-review & editing.

Funding

Not application.

Data availability

The whole-brain transcriptional profiles were obtained from the Allen Human Brain Atlas database (https://human.brainmap.org/static/download), and developmental specific gene expression datasets were obtained from the BrainSpan database (https://www.brainspan.org/static/download.html). The dysregulated gene expression of six major psychological disorders were derived from Gandal’s study (https://www.science.org/doi/10.1126/). Cellular specific marker gene-lists from five independent single-cell RNA-sequencing studies were summarized in Seidlitz’s study (https://www.science.org/doi/10.1126/). Cortical layer specific markers were obtained from He’s study (https://static-content.springer.com/esm/art%3A10.1038%2Fnn.4548). The code of MRI imaging preprocessing within the FreeSurfer software is found at http://surfer.nmr.mgh.harvard.edu, and the usage of python toolbox abagen is available at https://github.com/rmarkello/abagen. The comprehensive code for the construction of MIND networks was available on the github website: https://github.com/isebenius/MIND, and the code for PLS regression was available at https://github.com/SarahMorgan/Morphometric_Similarity_SZ. The Metascape enrichment analysis is available at http://metascape.org/, and CSEA analysis is available at https://genetics.wustl.edu/jdlab/csea-tool-2/. Other acquired datasets will be made available on request.

Declarations

Ethics approval and consent to participate

The protocol of this study was approved by the Ethics Committee of The Affiliated Yueqing Hospital of Wenzhou Medical University (ID: YQYY202500055), and was carried out in compliance with the Declaration of Helsinki. All written informed consents were acquired from all participants prior to the extensive screening procedure.

Consent for publication

Not applicable.

Competing interests

The authors declare that there are no competing interests.