Thalamocortical functional connectivity alterations and peripheral immune transcriptomic dysregulation in major depressive disorder patients with suicidal ideation

Abstract

Suicidal ideation (SI) in major depressive disorder (MDD) presents a serious clinical concern, yet its underlying biological mechanisms remain poorly understood. Using an exploratory, multimodal design, this study included 98 MDD patients with SI (MDD_SI), 61 without SI (MDD_nSI), and 233 healthy controls (HC), and collected functional MRI and peripheral blood transcriptomic data. Network-based statistics of resting-state functional connectivity (FC) identified brain network differences among the three groups, followed by graph-theoretical analysis of derived hub regions. Differential module connectivity (MDC) analyses of blood transcriptomic modules were performed to identify group-level differences, and exploratory correlation analyses were performed to examine associations between brain network topology and transcriptomic alterations. A functional network centered on the caudal temporal thalamus (cTtha) showed significant group differences (MDD_SI > MDD_nSI >HC; p < 0.001), with increased nodal efficiency of the left cTtha in both the MDD_SI (p = 0.043) and MDD_nSI (p < 0.001) groups compared with the HC group. Modules differentiating MDD_SI from MDD_nSI were enriched in antiviral immune responses and epigenetic regulation (darkred, MDC = 1.508, p = 0.04; yellow, MDC = 0.93, p = 0.01), while SI-specific alterations involved mitochondrial energy dysregulation and nSI-specific alterations involved vascular and translational pathways. Notably, one immune-related gene module was significantly correlated with left cTtha betweenness centrality (r = 0.274, p = 0.021), degree centrality (r = 0.262, p = 0.027), and nodal efficiency (r = 0.235, p = 0.048) in the MDD_SI group. This study indicates that MDD and SI are associated with cTtha-centered functional network alterations and peripheral transcriptomic dysregulation, and that, within MDD, SI may show a potential link between cTtha-centered brain network disruption and immune transcriptomic dysregulation. These results should be interpreted cautiously given the exploratory design, binary SI assessment, and modest effect sizes.

Introduction

Suicide is a major global public health concern, accounting for approximately 700,000 deaths annually worldwide, according to the World Health Organization (WHO). Suicidal ideation (SI), the initial stage of the suicide process, is a key predictor of future suicide attempts and deaths [1, 2]. Major depressive disorder (MDD), a leading contributor to the global burden of disease, accounted for an estimated 49.4 million disability-adjusted life-years (DALYs) in 2020 [3]. Individuals with MDD face an all-cause mortality rate approximately 1.6 times higher than that of the general population, with a nearly 20-fold increased risk of suicide [4]. In China, the lifetime prevalence of suicide attempts among individuals with MDD has been reported to reach 31% [5]. This elevated suicide risk substantially contributes to the overall disease burden associated with MDD [4]. Despite its clinical importance, the biological underpinnings of the SI remain poorly understood.

Over the past decade, a growing body of neuroimaging research has focused on identifying brain abnormalities associated with suicidality, aiming to establish objective neurobiological markers [6–8]. These studies have identified suicide-related alterations in both structure and function across multiple brain regions, including the medial prefrontal cortex, anterior cingulate cortex, hippocampus, thalamus, striatum, and insular cortex [6, 9, 10]. MDD patients with SI exhibit reduced connectivity within subcortical, salience, and ventral attention networks, as well as altered internetwork coordination involving ventral attention, frontoparietal control, and dorsal attention networks [11]. Additionally, hyperconnectivity within the frontoparietal network has also been observed in these patients [12]. Applying graph-theoretical analysis, the brain is conceptualized as a complex network of nodes and connections, and in MDD patients, nodal strength and local efficiency in the bilateral thalamus and left orbital superior frontal gyrus have been negatively correlated with SI [6]. The betweenness centrality of the left rostral middle frontal gyrus is also associated with SI [10]. These findings suggest alterations in both network integration and local information processing in these regions. However, inconsistency across studies persists, which probably stems from substantial heterogeneity in study design, including differences in sample characteristics, analytical methods, and age distributions. A critical limitation of the current research is the lack of studies that examine SI as a distinct construct [8, 13]. Many studies have combined individuals with SI and those with suicide attempts, despite increasing evidence suggesting that these two phenomena may involve partially dissociable neurobiological mechanisms [11, 14, 15]. Moreover, not all individuals who experience SI go on to attempt or die by suicide [1], further underscoring the necessity of identifying neurobiological markers specific to MDD patients with SI alone.

SI in MDD is a complex and multifactorial phenomenon that involves not only brain function dysfunction but also systemic biological dysregulation [16]. Accumulating evidence suggests that a variety of peripheral biological processes may contribute to the pathophysiology of suicide in MDD, including neuroimmune dysregulation [17], inflammatory activity [18], protein misfolding and endoplasmic reticulum stress [19], metabolic disturbances [20], and impaired energy metabolism [21]. MDD is increasingly conceptualized as a condition of chronic low-grade inflammation, characterized by immune activation, increased proinflammatory cytokines, and dysregulated inflammatory signaling pathways [22]. However, given that no single biomarker can adequately explain the heterogeneity of MDD, transcriptomic profiling has emerged as a powerful data-driven approach for characterizing gene expression alterations associated with complex psychiatric phenotypes [23]. Compared with postmortem brain tissue, peripheral blood samples are more readily accessible, can dynamically reflect disease states, and show a high degree of concordance with brain gene expression patterns, making them a reliable window into central processes in living individuals [23–25]. By integrating transcriptomic data with neuroimaging measures, it becomes possible to directly investigate how peripheral molecular signatures relate to specific alterations in brain function, thereby elucidating the biological mechanisms of SI at a system level. Moreover, to further interpret the molecular data, weighted gene co-expression network analysis (WGCNA) offers robust statistical tools for identifying disease-specific molecular profiles [26]. As such, data-driven peripheral transcriptomic analysis has emerged as a promising approach for identifying molecular modules from high-dimensional data that are specific to MDD patients with SI, with the potential to uncover clinically relevant biomarkers.

This study aims to offer an exploratory, integrative analysis of potential neural and molecular mechanisms associated with SI in individuals with MDD. First, we utilized the network-based statistics (NBS) approach to identify SI-related functional brain subnetworks and candidate hub regions. These NBS-derived hubs were then examined using graph-theoretical measures, including degree, betweenness centrality, and nodal efficiency, to characterize their roles in network connectedness, information flow, and integration [10, 27]. Second, WGCNA was performed to detect gene co-expressed modules significantly associated with SI on the basis of peripheral blood transcriptomic data [28]. Functional enrichment analysis was subsequently conducted to interpret the biological significance of these gene modules. Finally, we examined the potential associations between SI-related brain functional alterations and gene expression patterns (Fig. 1). By integrating neuroimaging and transcriptomic data, this study seeks to provide a cross-level understanding of the biological mechanisms contributing to SI in depression and to identify candidate biomarkers for future clinical practice.

Fig. 1. Schematic representation of the study.

A Static functional neuroimaging analysis, including neuroimaging preprocessing, functional connectivity network construction, network-based statistical analysis, and topological characterization of hub nodes. B Transcriptional profile analysis, including weighted gene co-expression network analysis, module differential connectivity analysis, and functional enrichment analysis. C Correlation analysis between functional connectivity networks and transcriptional co-expression modules. MDD, major depressive disorder; SI, suicidal ideation; ROI, region of interest; LOC, loss of connectivity; GOC, gain of connectivity.

Materials and methods

Subjects

A total of 159 patients were recruited from the Mental Health Center of West China Hospital of Sichuan University between January 2016 and December 2020. None of the participants had received psychotropic medications within the three months prior to study enrollment. The inclusion criteria were as follows: aged between 16 and 60 years; currently experiencing a major depressive episode, confirmed by the Diagnostic and Statistical Manual of Mental Disorders Fourth Edition (DSM-IV) criteria for MDD; and a score of ≥17 on the 17-item Hamilton Depression Rating Scale (HAMD) [29]. The exclusion criteria included any other DSM-IV Axis I/II disorders, neurological diseases, brain injury, or history of suicide attempts.

Additionally, 233 healthy volunteers with no history of psychiatric disease were recruited via public advertisements. Individuals with a family history of psychiatric disorders across two lineages within three generations, any neurological conditions, or unstable physical health were excluded. The study was approved by the Institutional Ethics Committee of West China Hospital, Sichuan University (Approval Number: [2016] 170), and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants, and from legal guardians for those under 18 years of age. The recruitment procedure is illustrated in Figure S1.

Assessment of suicidal ideation

SI was assessed through clinical interviews conducted by trained psychiatrists as part of the standardized psychiatric evaluation. Patients were directly asked the standardized question: “Since the onset of illness, have you thought about suicide?” Based on their responses, patients who answered “yes” were classified into the SI group (n = 98), whereas those who answered “no” were classified into the non-SI (nSI) group (n = 61) [30]. To further reduce the potential confounding effects of suicidality on overall symptom severity, we computed an adjusted HAMD total score by subtracting the score of item 3 (suicidality) from the HAMD total score [7].

Acquisition of neuroimaging data and topological analysis of the brain functional network

All participants completed neuroimaging assessments, including resting-state functional MRI and high-resolution T1-weighted anatomical scans. The parameters for MRI data acquisition and the preprocessing procedure are detailed in the “Methods” section of the Supplementary Materials. Each network node was derived from a region of interest (ROI) defined in the Brainnetome Atlas, which segments the brain into 210 cortical and 36 subcortical regions [31]. Extracted mean time series for each ROI and computed Pearson correlations, yielding a 246 × 246 connectivity matrix.

To explore group-level differences in FC networks, the NBS toolbox [27] was utilized. One-way analysis of variance (ANOVA) was conducted at the edge level to compare network connectivity among the MDD_SI, MDD_nSI, and HC groups. The analysis parameters were defined as follows: a primary F statistic threshold of 10 was applied for the NBS, with statistical significance assessed using 10,000 random permutations. Within the significant subnetwork identified by NBS, hub regions were defined as nodes exhibiting the highest number of connections [6]. Group differences were assessed at both the subnetwork and edge levels. Mean subnetwork FC was compared using linear models with age, sex, and framewise displacement (FD) included as covariates, followed by post hoc pairwise tests, and individual connections within the subnetwork were further examined using post hoc pairwise comparisons.

Network topological measures were computed via MATLAB routines implemented in the GRETNA toolbox [32]. Betweenness centrality, degree centrality, and nodal efficiency were used to characterize the hub topology within the subnetwork, reflecting complementary aspects of information mediation, local connectedness, and communication efficiency [6, 10].

Peripheral blood RNA collection, data processing, and the constructed gene co-expression module

RNA was extracted from peripheral whole blood was extracted and sequenced via standard protocols. Data preprocessing, alignment, and normalization followed established pipelines. Low-expression transcripts were filtered (see the Supplementary Materials for full experimental details). Gene co-expression module and module preservation analyses were performed via the WGCNA package in R [28], which allows the identification of biologically meaningful gene modules based on expression correlation patterns and facilitates the assessment of module preservation across diagnostic groups. The HC group was used as the reference since it represents a biologically stable baseline for identifying disease-related perturbations in network organization [33]. RNAs exhibiting high expression correlations in the reference set were clustered into co-expression modules, which were assigned arbitrary names. Each module represents a group of highly interconnected genes characterized by strong absolute correlations. Co-expression modules were constructed with a minimum module size of 30 and a merging height cutoff of 0.1.

Module preservation statistics were applied to assess whether the identified modules were preserved across the MDD_SI, MDD_nSI, and HC groups. Preservation metrics evaluate the extent to which within-module connectivity patterns are retained in independent datasets. In this study, a composite statistic (Z_summary) was used, which combines density-based and connectivity-based preservation measures. A Z_summary > 10 indicates strong evidence of preservation, 2–10 suggests weak to moderate preservation, and < 2 indicates no evidence of preservation [33].

Module differential connectivity

Module differential connectivity (MDC) analysis was conducted via the DGCA package [34] in R to assess differences in gene co-expression modules between groups (MDD_SI vs. MDD_nSI, MDD_SI vs. HC, MDD_nSI vs. HC), controlling for age and sex as covariates. This approach compares the connectivity ratio of all gene pairs within a module between groups, providing a quantitative measure of intramodular connectivity changes. An MDC value > 1 indicates gain of connectivity (GOC) or enhanced gene co-expression, whereas values < 1 reflect loss of connectivity (LOC) or reduced co-expression among genes [26, 35, 36].

Functional enrichment analysis and hub genes

Gene Ontology (GO) enrichment analysis was performed to investigate the biological functions of genes within the identified modules via the clusterProfiler package [37] in R (v4.1.0). Biological process (BP) terms with false discovery rate (FDR)-corrected p values < 0.05 were considered significantly enriched. For each module, the top 10 BP terms with the lowest adjusted p values were retained for functional annotation. Hub genes with the top five intramodular connectivity were selected.

Differential expression and integration with WGCNA modules

Differentially expressed gene (DEG) analysis among the three groups was performed using the limma package (v3.21) in R. A linear model was fitted with diagnostic group as the main factor, adjusting for age and sex as covariates. DEGs were defined using thresholds: pFDR < 0.05 and |log₂ fold change| ≥ 1.2. Post hoc pairwise comparisons between groups were conducted using limma contrasts. DEGs were further mapped to WGCNA co-expression modules based on ME, and the enrichment of DEGs within each module was evaluated using a hypergeometric test. The association between log₂ fold change and ME was examined to evaluate the distribution of transcriptional perturbations within modules.

Statistical analysis

Normality and homogeneity of variance were assessed prior to parametric statistical analyses. The chi-square test was used for categorical variables (sex and first onset). Continuous variables (age, FD, mean FC, and graph theory metrics) were compared across the MDD_SI, MDD_nSI, and HC groups using one-way analysis of variance (ANOVA) and are presented as mean ± standard deviation (SD). Differences in total disease duration between the patient subgroups were tested via Student’s t test. The Wilcoxon test was applied to compare HAMD scores between the patient subgroups. Spearman’s correlation analysis was performed to explore the relationships between the brain network, graph theory metrics and gene co-expression modules, controlling for age, sex, and FD as covariates. Statistical significance was defined as a two-tailed p value < 0.05, and all results were corrected for multiple comparisons via the FDR method. All analyses were conducted using R software (v4.1.0). The custom scripts used in this study are available from the corresponding author upon reasonable request.

Results

Demographic and clinical characteristics

The participants’ demographic and clinical characteristics are presented in Table 1. No significant differences were found in sex, age or FD across the three groups. Patients in the SI group had higher HAMD-3 subscore, and longer durations of current depressive episodes than those in the non-SI group. No significant differences were found between the two patient groups in terms of first episode, total duration, total HAMD scores, or adjusted HAMD scores.

Table 1.

Demographic and clinical information of participants.

Characteristics	Health controls (n = 233)	MDD without suicidal ideation (n = 61)	MDD with suicidal ideation (n = 98)	Statistics	p
Age (years)	26.93 ± 9.7	30.26 ± 10.11	27.43 ± 9.08	2.924 a	0.055
Sex (Male/Female)	76/157	25/36	38/62	1.916 b	0.384
Head motion (mm)	0.07 ± 0.03	0.07 ± 0.02	0.08 ± 0.046	0.546 a	0.58
First onset (Yes/No)	NA	41/20	58/40	0.888 b	0.376
Total duration (months)	NA	30.46 ± 59.68	33.81 ± 45.70	0.395 c	0.693
Duration of current episode (months)	NA	4.03 ± 2.70	8.09 ± 14.62	2.103 c	0.037
HAMD total score	NA	20.23 ± 4.42	21.63 ± 5.32	1.724 c	0.087
HAMD-3 item score	NA	0.72 ± 0.78	1.61 ± 0.93	6.243 c	<0.001
Adjusted HAMD total score	NA	19.51 ± 4.19	20.02 ± 4.85	0.683 c	0.496

MDD, major depressive disorder; HAMD, Hamilton depression rating scale.

^aOne-way ANOVA.

^bChi-square test.

^cStudent’s t-statistics.

Group differences in the brain network and local graph metrics of the hub region

NBS analysis revealed a subnetwork with significant differences in edge weight across the three groups (Table S2, Fig. 2A–C). The identified subnetwork consisted of 52 nodes and 64 edges, with key nodes predominantly located in the thalamus and frontal cortex, notably the left caudal temporal thalamus (cTtha, 38 edges), right cTtha (9 edges), and bilateral lateral superior frontal gyrus (SFG, 8 edges). Additional regions were distributed across several canonical functional networks, including the frontoparietal network (FPN; inferior middle frontal gyrus [MFG], rostrodorsal inferior parietal gyrus [IPG], caudolateral inferior temporal gyrus [ITG]), default mode network (DMN; lateral and medial SFG, caudal and rostral middle temporal gyrus [MTG]), and subcortical network (SC; dorsal caudate, medial prefrontal thalamus [mPFtha], rostral temporal thalamus [rTtha], occipital thalamus [Otha]). Post hoc analyses revealed significant differences in the mean FC of the subnetwork among the three groups (MDD_SI vs. HC, p < 0.001; MDD_nSI vs. HC, p = 0.021; MDD_SI vs. MDD_nSI, p = 0.032) (Fig. 2D). Edge-level post hoc analyses further revealed that, compared with the HC group, both the MDD_nSI and MDD_SI groups exhibited significantly increased connectivity between the cTtha and DMN networks as well as within the frontoparietal network, with the alterations mainly involving prefrontal and inferior temporal regions. In addition, multiple thalamocortical edges differed between the MDD subgroups, involving regions within the visual network, sensorimotor network, and limbic system (Table S3, Figure S2).

Fig. 2. Comparison of brain network characteristics and topological metrics across the MDD_SI, MDD_nSI, and HC groups.

A A functional network containing 52 nodes and 64 edges shows group differences among the MDD_SI, MDD_nSI, and HC groups. B Circular diagram of the functional network. C The average glass brain connectome for each group, showing edgewise connectivity patterns. D Differences in network-averaged functional connectivity among the MDD_SI, MDD_nSI, and HC groups are illustrated via a cloudrain plot. E Comparative analysis of local topological metrics of the left cTtha across the three groups. Only significant differences are indicated. HC, healthy control; MDD_SI, major depressive disorder with suicidal ideation; MDD_nSI, major depressive disorder without suicidal ideation; SC, subcortex; VIS, visual network; SMN, somatomotor network; DAN, dorsal attention network; VAN, ventral attention network; LIM, limbic network; FPN, frontoparietal network; DMN, default mode network.

The left cTtha showed the highest number of connections within the subnetwork and was therefore identified as the hub region of this subnetwork (Fig. 2A, B). Significant group differences were observed in local graph-theoretical metrics in the left cTtha region, including degree centrality (F = 11.14, p < 0.001) and nodal efficiency (F = 10.68, p < 0.001), whereas no significant group difference was found for betweenness centrality (F = 0.13, p = 0.88). Post hoc comparisons revealed that degree centrality and nodal efficiency were greater in both MDD_SI (p = 0.048) and MDD_nSI (p < 0.001) groups than in the HC group (Fig. 2E).

Group differences in gene co-expression modules

Transcriptome sequencing was completed for 208 participants, resulting in the alignment of 61,806 mRNA transcripts to the reference genome (Table S1). After the lowly expressed transcripts were removed, 11,722 genes were retained for downstream analysis. Co-expression module construction across the three groups was performed using a soft-thresholding power of 6, which was chosen to approximate scale-free topology. This process resulted in 23 distinct gene modules, each ranging from 36 to 3223 genes, which were uniquely color-coded for identification. Transcripts that could not be assigned to any specific module were grouped into a gray module, encompassing 580 mRNAs (Table S4). Assessment of module preservation revealed that the overall transcriptional structures were stable in both the MDD_nSI and MDD_SI groups, with all modules showing Z summary scores exceeding 2 (Figure S3).

Group-level differences in module connectivity were subsequently analyzed using the MDC metric (Table 2). Significant differences were observed between the MDD_SI and MDD_nSI groups in the darkred module (MDC = 1.508, p = 0.04) and yellow module (MDC = 0.93, p = 0.01). Comparisons between the MDD_nSI patients and HCs revealed significant differences in the lightcyan module (MDC = 1.09, p = 0.01), brown module (MDC = −0.104, p = 0.03), and royalblue module (MDC = −0.979, p = 0.03). Significant differences between the MDD_SI patients and HC were identified in the royalblue module (MDC = 0.954, p = 0.02) and turquoise module (MDC = 0.226, p = 0.04).

Table 2.

Comparison of differences in module connectivity across groups.

Module	size	MDC value	p	comparisons
darkred	39	1.508	0.04	MDD_SI vs. MDD_nSI
yellow	1022	0.93	0.01	MDD_SI vs. MDD_nSI
lightcyan	79	1.09	0.01	MDD_nSI vs. HC
brown	1483	−0.104	0.03	MDD_nSI vs. HC
royalblue	41	0.979	0.03	MDD_nSI vs. HC
		0.954	0.02	MDD_SI vs. HC
turquoise	3223	0.226	0.04	MDD_SI vs. HC

MDC, module differential connectivity; MDD_SI, major depressive disorder with suicidal ideation; MDD_nSI, major depressive disorder without suicidal ideation; HC, healthy control.

At the module level, relative to the MDD_nSI group, the MDD_SI group presented increased connectivity in the yellow module (4.3%) and decreased connectivity in the darkred module (4.3%), with no significant changes observed in the remaining 21 modules (91.4%). Compared with HCs, MDD_SI patients presented decreased connectivity in the royalblue and turquoise modules (8.7%) and connectivity in the other 21 modules (91.3%) remained unchanged. Similarly, the MDD_nSI group displayed decreased connectivity in the brown and royalblue modules (8.6%) and increased connectivity in the lightcyan module (4.3%), with no significant differences detected in the remaining 20 modules (87.0%).

Functional annotation and hub genes of the co-expression modules

The darkred module was associated with antiviral-related program, encompassing host responses to viral infection, regulation of viral replication and entry, and interactions with symbiotic organisms, with hub genes related to interferon signaling and pathogen restriction (IRF7, IFI35 and IFITM3). The yellow module was primarily enriched in protein posttranslational modification, chromatin remodeling, and DNA repair mechanisms and was driven by epigenetic regulators and ubiquitin-associated genes such as HECTD1, CASP8AP2, AKAP9, PHC3 and MYCBP2. The lightcyan module reflects key pathways involved in hemostasis and vascular stability, including coagulation, platelet activation, and wound repair, with platelet-associated genes (MPIG6B, CTDSPL, SPARC and GP1BA) acting as central network hubs. The brown module was enriched for essential processes in protein synthesis and mitochondrial energy metabolism, including ribosome biogenesis, translation, and oxidative phosphorylation, with hub genes dominated by ribosomal proteins and respiratory chain components (RPL6P27, TRAF7, COX7C and UFC1). The royalblue module was linked to lipid handling and steroid hormone regulation, encompassing aldosterone biosynthesis, lipid transport, and synaptic vesicle endocytosis, with hub genes including PRKAR2B, CAVIN2, PDE5A, ELOVL7 and GUCY1B1. The largest turquoise module highlighted mitochondrial energy metabolism and nucleotide triphosphate turnover, including oxidative phosphorylation, ATP synthesis, and cellular respiration, and was centered on mitochondrial electron-transport and ATP-synthase genes such as MCRIP1, IDH3G, FAM50A, PLEKHJ1 and FBXL15 (Fig. 3, Table S5–6).

Fig. 3. Top ten enriched biological process terms and hub genes of the modules.

A–F display the gene co-expression modules: darkred, yellow, lightcyan, brown, royalblue, and turquoise, respectively.

Differential expression and integration with co-expression modules

A total of 138 DEGs were identified across HC, MDD_nSI, and MDD_SI groups and mapped to co-expression modules (Table S7). Most DEGs were enriched in the yellow (n = 41), brown (n = 14), lightcyan (n = 6), and turquoise (n = 10) modules, showing distinct regulatory patterns between MDD_SI and MDD_nSI. Specifically, DEGs in the yellow module were predominantly upregulated in MDD-SI, whereas the lightcyan module showed uniform downregulation, and the brown and turquoise modules exhibited mixed expression changes. Significant enrichment was observed in the yellow (n = 40) and turquoise (n = 10) modules (Table S8). The brown, turquoise, yellow, and lightcyan modules showed significant core gene–driven expression changes with opposite correlation directions between MDD_nSI vs. HC and MDD_nSI vs. MDD_SI (|r| = 0.50–0.85, all p < 0.05). The darkred module showed moderate correlation in HC vs. MDD_nSI (r = 0.57, p = 0.00014) with a reversed direction in MDD_nSI vs. MDD_SI (r = −0.47, p = 0.0026), whereas no significant correlations were observed for the royalblue module (all p > 0.05) (Table S9).

Correlations between brain network metrics and gene co-expression modules

Exploratory correlation analyses were conducted in the MDD_SI group, and the following associations are reported without correction for multiple comparisons. Within this context, the darkred module was positively correlated with betweenness centrality (r = 0.274, p = 0.021), degree centrality (r = 0.262, p = 0.027), and nodal efficiency (r = 0.235, p = 0.048) in the left cTtha region (Table S10, Fig. 4). The yellow module was negatively associated with betweenness centrality (r = −0.311, p = 0.008). The mean FC of the brain network (r = 0.23, p = 0.021), degree centrality (r = 0.214, p = 0.033) and nodal efficiency (r = 0.242, p = 0.015) of the left cTtha region were correlated with adjusted HAMD scores in the MDD_SI group.

Fig. 4. Correlations between the cTtha-centered brain network, gene co-expression modules, and depressive symptoms in the MDD_SI group.

cTtha, caudal temporal thalamus; MDD_SI: major depressive disorder with suicidal ideation; HAMD, Hamilton Depression Rating Scale; Adjusted HAMD total score = HAMD total score – HAMD-3 score. * p < 0.05; ** p < 0.01.

Discussion

This multimodal analysis indicates that most brain network and transcriptomic alterations are shared between MDD patients with and without SI, as reflected by convergent thalamocortical network abnormalities and largely preserved gene co-expression architectures, supporting a largely common neurobiological basis of depression. Both patient groups showed convergent thalamocortical network abnormalities and broadly preserved gene co-expression architectures, whereas SI-related differences were confined to a limited number of connections and modules. Notably, MDD_nSI exhibited pronounced but reversible changes in core gene regulation, which may reflect a transitional regulatory state. In contrast, SI was associated with selective reorganization of specific modules, particularly those related to immune and epigenetic processes.

Our findings indicate that, despite comparable overall depressive severity and illness duration, patients with SI exhibited a distinct clinical profile characterized by selectively elevated HAMD-3 subscore indexing guilt, suicidal thoughts, and hopelessness, along with a longer duration of the current depressive episode. This pattern supports previous evidence that global depression scores may fail to capture clinically meaningful features specifically related to SI [16]. Although suicidality has been reported to be more prevalent in the early course of mood disorders [38], the prolonged current episode observed in our SI group may reflect the chronic and persistent nature of depressive symptoms associated with SI.

At the brain network level, we observed marked hyperconnectivity within a left cTtha-centered brain subnetwork in MDD, consistent with accumulating evidence for excessive thalamocortical integration in depression [39]. The thalamus, far beyond a mere relay station, occupies a central position in coordinating cortico-cortical communication, regulating information flow, and synchronizing activity across distributed functional networks, transmitting sensory information to the cortex, modulating cortical activity, and integrating motor-related signals essential for perception, cognition, and behavior [40]. Disruptions of ascending sensory-integration and descending cognitive-modulation thalamocortical loops have been implicated in affective instability, altered salience attribution, and impaired cognitive flexibility in MDD [41]. Furthermore, patients with SI exhibited a further increase in mean FC within this subnetwork compared with their non-SI counterparts, suggesting an amplification of a shared MDD-related network rather than a qualitative reorganization. Within the context of MDD, such amplification may reflect increased intrinsic signal gain or reduced network selectivity, potentially facilitating the persistence of aversive internal states such as psychological pain, a well-established proximal driver of SI [42]. Other thalamic nuclei, including mPFtha, rTtha, and Otha, were also involved in the network. Given the functional heterogeneity of thalamic nuclei, further research is warranted to elucidate the specific role of cTha in the pathophysiology of MDD and suicidality.

Beyond the thalamus, network alterations extended into higher-order cognitive control and self-referential systems. In particular, regions within the FPN and DMN were embedded within the thalamus-centered subnetwork. As a core system for executive function, the FPN centers on major hubs, including the MFG, IPL, and ITG, which are extensively connected to cortical regions that support attention, decision-making under uncertainty, and complex problem solving [43, 44]. Altered connectivity within the FPN has been linked to suicide, suggesting that this network may contribute to behavioral disinhibition and impaired decision-making, both of which are recognized as risk factors for SI [45, 46]. Disruptions in the FPN, particularly those affecting its role in impulse control, may impair the evaluation of behavioral consequences, thereby increasing suicide risk [47]. In support of this, individuals at elevated suicide risk exhibit abnormal FPN connectivity patterns [12]. Our results indicate that aberrant FPN connectivity may represent a potential neural substrate underlying impaired inhibitory control and heightened impulsivity in individuals with MDD and SI.

In addition, network alterations encompassed key components of the DMN, including the lateral/medial SFG and caudal/rostral MTG, which play a role in consciousness, self-referential processing, social cognition, and autobiographical memory. This result aligns with evidence that SI is closely linked to heightened rumination, perceived entrapment, and an inability to shift attention away from internally generated distress [48]. A neuroimaging review suggested that altered functioning in the lateral/medial SFG and affective-reward circuits may contribute to SI by promoting heightened negative affect and blunted responses to positive internal states [9]. Furthermore, increased FC between the ventromedial prefrontal cortex and medial SFG has been reported to be specific to individuals with MDD who have attempted suicide [49]. In line with this, parcellation-based analyses of the orbital and anterior medial prefrontal cortex have demonstrated that the medial SFG is closely involved in affective and self-referential processing and has been implicated in the pathophysiology of mood disorders [50]. These findings suggest that SI reflects an interaction between compromised executive control and amplified self-referential processing within a shared thalamocortical framework, with the thalamus—particularly the cTtha—serving as a central hub mediating the failure to regulate maladaptive DMN activity.

At the peripheral molecular level, enrichment of antiviral and type I interferon-related immune processes in the darkred module, including interferon-stimulated hub genes (IRF7, IFI35, and IFITM3), points to immune dysregulation shared across MDD subgroups and aligns with the view of MDD as a disorder characterized by chronic low-grade inflammation [51]. Several published studies provide compelling evidence supporting that exposure to infections and immune activation are associated with an increased risk of MDD and suicide. The Danish Blood Donor Study revealed that herpes simplex virus type 1 (HSV-1) infection increased the risk of suicide-related behavior by 1.4-fold [52]. Accumulating evidence indicates that both patients with MDD and those with SI exhibit abnormal peripheral levels of multiple inflammatory mediators, including IL-6, CRP, and TNF-α, which are consistent with a dysregulated immune response observed in both conditions [51, 53, 54]. Transcriptomic analysis of peripheral blood indicated that MDD patients, particularly those with SI, exhibit dysregulated immune and viral response pathways, including HSV-1 infection, NOD-like receptor signaling, and antigen processing and presentation, as well as activation of innate and adaptive immune pathways such as TLR and NF-κB signaling, inflammasome activity, and monocyte-associated inflammation [23, 55].

Our findings also suggest the relative prominence of dysregulation in mitochondrial energy metabolism and lipid-related pathways in the MDD_SI group, supported by the involvement of genes such as IDH3G, a core component of the tricarboxylic acid cycle, and ELOVL7, which participates in long-chain fatty acid metabolism. Mitochondria play a central role in regulating ATP production, lipid metabolism, and protein synthesis. Growing evidence suggests that disruptions in mitochondrial energy metabolism are closely linked to the pathophysiology of MDD and suicide [56, 57]. It has been proposed that the therapeutic effects of antidepressants on depression-like behavior may be mediated through the activation of mitochondrial respiratory chain function and subsequent enhancement of ATP synthesis [58]. A transcriptomic analysis of postmortem dorsolateral prefrontal cortex tissue revealed altered mitochondrial metabolism in individuals who died by violent suicide [59], and changes in gene expression related to ATP biosynthesis have also been observed in suicide completers [60]. Furthermore, patients with MDD exhibit disrupted lipid metabolism, characterized by decreased levels of long-chain monounsaturated and saturated fatty acids and increased levels of lysophospholipids [61]. A 12-month longitudinal study of antidepressant treatment suggested that the combined assessment of cortisol, total cholesterol, and folate may serve as a predictive marker for suicide attempts [62]. Although findings remain somewhat controversial, the majority of studies suggest that supplementation with polyunsaturated fatty acids may be beneficial for alleviating depressive symptoms and reducing suicide risk [63]. Thus, these observations underscore the critical interplay between mitochondrial function, lipid metabolism, and neurobiological vulnerability in the etiology of MDD and suicide.

In addition, transcriptional alterations observed in the MDD_nSI group were predominantly enriched for pathways related to hemostasis and vascular stability, and platelet function, consistent with the involvement of key platelet adhesion and activation regulators such as GP1BA and MPIG6B. Beyond their classical role in coagulation, platelets are increasingly recognized as active participants in immune regulation through the secretion of proinflammatory and immunomodulatory mediators [64]. Psychological stress has been linked to increased and prolonged proinflammatory platelet bioactivity, further implicating platelet dysfunction in the pathophysiology of depression [65]. Blood platelets may serve as potential biomarkers for the prediction and progression of MDD, especially in the context of comorbid conditions [66]. Associations have also been reported between platelet count and insomnia and restlessness dimensions of depressive symptoms [67]. Therefore, transcriptional alterations of platelets appear to index depressive symptom burden, whereas SI is distinguished by immune–metabolic dysregulation.

Finally, the integration of transcriptomic and neuroimaging findings revealed that immune and antiviral transcriptional modules in the MDD_SI group were associated with a thalamus-centered brain network implicated in SI. Prior studies have indicated that immune and inflammatory dysregulation can impair the structure and function of key brain regions, such as the prefrontal cortex and limbic system, which are crucial for emotional regulation [68, 69]. Cytokine imbalances have been associated with thalamocortical dysfunction, potentially contributing to mood instability and SI [70, 71]. These exploratory findings tentatively support a framework in which immune dysregulation is associated with thalamocortical network organization, together suggesting a possible multisystem biological context for elevated suicide risk in MDD that requires confirmation in future rigorously designed studies.

Limitations

There were several limitations in our study. First, SI was assessed via a single dichotomous interview-based item, which did not capture variations in SI severity, frequency, or persistence. Future studies should employ validated, dimensional SI-specific instruments. Second, due to sample size limitations, individuals with prior suicide attempts were excluded, which limits the applicability of the findings across the full spectrum of suicidal behavior, and longitudinal studies are warranted to explore the transition from SI to behavior. Third, the lack of clinical immunobiomarkers and immunological history data limited the ability to investigate immune-related mechanisms underlying the observed findings. Fourth, the differences between the SI and non-SI groups were relatively modest, which may constrain the strength of SI-specific interpretations. Fifth, the clinical variable named episode duration differed between patient subgroups and may have contributed to the observed patterns; their potential confounding effects cannot be fully excluded. Finally, as this study was exploratory in nature, the imaging-gene correlations did not survive multiple correction, which may limit the robustness and generalizability of the findings.

Conclusion

In conclusion, this exploratory, cross-sectional study highlighted novel associations between thalamus-centered brain network properties and peripheral immune-related transcriptomic pathways in individuals with MDD and SI. These findings provide valuable insight into the potential interplay between central neural circuits and peripheral immune processes in SI. Further longitudinal and SI-targeted studies are essential to validate these findings, elucidate the underlying mechanisms, and assess their clinical relevance for suicide prevention.

Author contributions

MW, JXW, XY and XHM conceived and designed the study. MW and JXW performed the data analysis. MW, JXW, YSY, YKD, LSZ and RJN collected the data. MW drafted the manuscript. XY and XHM supervised the study. All authors contributed to revising the manuscript and approved the final version.

Funding

This work was supported by the Ministry of Science and Technology of the People’s Republic of China (grant number 2022ZD0211700), the 135 Project from West China Hospital of Sichuan University (grant numbers 2023HXFH006, 2023HXFH040), and the Postdoctoral Research Fund of West China Hospital, Sichuan University (grant number 2024HXBH135).

Data availability

The RNA sequencing data generated in this study are available in the Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA009854). Other raw data supporting the findings of this study are available from the corresponding author upon reasonable request.

Ethics declarations

This study was approved by the Institutional Ethics Committee of West China Hospital, Sichuan University (Approval Number: [2016] 170), and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04047-w.