This cross-sectional study is part of a project aimed at evaluating biomarkers in patients maintained on APs in Qatar. Refer to other published papers from this project for more details on the methodology (Hammoudeh et al., 2018; Engelke et al., 2022; Kiwan et al., 2019). The institutional review boards (IRB) at Hamad Medical Corporation (HMC; number 16-0002) and Weill Cornell Medicine in Qatar (WCMQ; number 1056791-10) approved the study. Patients (BP and SSD subjects) were recruited from Qatar’s Mental Health Services, and the control group was recruited from primary care clinics and visitors to the Mental Health Hospital. All participants provided written informed consent before enrollment. The presence or absence of a psychiatric disorder was confirmed using the standardized Mini International Neuropsychiatric Interview (Sheehan et al., 1998), which follows the Diagnostic and Statistical Manual (DSM) criteria for diagnosing psychiatric disorders.

Briefly, the inclusion criteria for patients were ages 18–65 years and at least 6 months of maintenance on APs. Patients in the MD + APs group were treated with various antipsychotic medications, including both FGA and SGA. Those not on APs met the same criteria, except they had been off APs for at least 6 months. We did not recruit first-episode subjects or participants completely naïve to psychotropics. The inclusion criteria for healthy controls (HC) included the same age range, but they had no mental illness or AP use. For this study, we selected a total of 64 patients diagnosed with SSD and BP, both on and off APs, along with 92 HC matched for age and gender.

Trained raters collected demographic and clinical data, including age, sex, and APs taken. They also measured body mass index (BMI), waist circumference, and blood pressure (systolic and diastolic). Fasting blood samples were collected (typically in the morning after 10–12 h of fasting) within 1–2 days of recruitment and stored for subsequent testing. Metabolic measures, including lipid profile (cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL), glycosylated hemoglobin, and fasting blood glucose, were measured directly from these samples using the same standardized procedures employed at Hamad General Hospital, Doha, Qatar.

Paxgene Blood RNA (Qiagen Catalog no. 762164) kit was used to extract RNA from peripheral blood, and Qubit was used to determine the RNA concentration. To ensure high-quality RNA suitable for transcriptome profiling, rRNA and globin transcripts were depleted using Lexogen’s RiboCop rRNA Depletion Kit for Human/Mouse/Rat plus Globin. After depletion, stranded RNA-seq libraries were prepared using Lexogen’s CORALL RNA-Seq V2 Kit, which preserves strand orientation to enhance transcript resolution. All samples met the library quality control standards prior to sequencing. The libraries were then sequenced on the Illumina NovaSeq 6000 platform using the S4 v1.5 200-cycle configuration, producing high-throughput paired-end reads for downstream expression analysis. Sequencing summary statistics, including read depth and CG%, are provided in Supplementary Table S1. All samples contain over 61 million reads, with an average Q score of 35.

Transcript-level abundance estimates were obtained using the Illumina DRAGEN Bio-IT Platform, which generated quant. sf files for each sample via Salmon (Patro et al., 2017). The tximport R package was used to summarize transcript-level estimates into gene-level counts (Soneson et al., 2015), applying a transcript-to-gene mapping derived from the GENCODE v36 GTF annotation. These gene-level counts were then used to construct a DESeqDataSet object for differential expression analysis with DESeq2 (Love et al., 2014). Quality control steps included filtering out genes with counts <10, and no outlier samples were identified using hierarchical clustering and principal component analysis (PCA). Differential gene expression analysis was performed using the DESeq2 package (v1.46.0), with adjustment for gender and a significance threshold of an adjusted p-value (FDR) < 0.05 and a log2 fold change cutoff of 1.5. MA plots were generated using “lfcShrink” for all contrasts to assess dispersion patterns and potential biases caused by group size imbalance (Supplementary Figure S1). The PCA plot was generated using PCA Explorer (Bioconductor) based on variance-stabilized RNA-seq counts (VST) obtained from the DESeq2 pipeline after gene filtering.

Functional enrichment was then performed using the Metascape tool to identify specific altered pathways (Zhou et al., 2019). First, we compared mental disorder off APs (MD) and mental disorder on APs (MD + APs) versus HC to identify differentially expressed genes (DEGs). Subsequently, we categorized the groups by disease and compared each of the following groups: SSD, SSD + APs, BP, and BP + APs, independently vs. the HC. We then compared the identified DEGs with genes previously reported to be associated with SSD and BP in independent studies: (Pan et al., 2025; Wang et al., 2024; Cui et al., 2021; Hess et al., 2020; Krebs et al., 2020; Munkholm et al., 2015).

The demographic, clinical, and metabolic data were analyzed using SPSS (version 29), with ANOVA for continuous variables and Chi-Square tests for categorical variables to assess differences among the three groups. The p-value was set at 0.05, and Bonferroni corrections were applied to multiple post hoc comparisons, as specified in SPSS.

The main goal of network co-expression analysis is to identify biologically and clinically relevant modules and genes. A scale-free co-expression network was built using soft thresholding. The optimal soft threshold was selected to cluster this network into different functional modules. Modules were hierarchically clustered by calculating the eigengene, and similar modules were merged. Then, we ran a linear model to determine which modules contribute to differences across groups. We also correlated the modules identified through WGCNA with clinical metabolic features.

Protein–protein interaction (PPI) network analysis was conducted using Metascape, an online platform (http://metascape.org). The list of differentially expressed genes (DEGs) was uploaded into Metascape for functional annotation and interaction analysis. PPI networks were created by integrating data from publicly available databases: STRING, BioGrid, OmniPath, and InWeb IM (Li et al., 2017; Stark et al., 2006; Szklarczyk et al., 2019).

To explore the transcriptomic signatures in individuals with mental illness, we first performed a differential expression analysis comparing all individuals with mental illness to healthy controls (HC), regardless of antipsychotic treatment status. This analysis identified 1,565 upregulated genes and 276 downregulated genes (Supplementary Table S2). Functional enrichment analysis of the differentially expressed genes revealed pathways related to immune activation and immune cell differentiation, including positive regulation of immune response, T-cell activation, and selection. Additionally, enriched pathways involved RNA metabolism and translation, as well as DNA metabolic and repair processes. These findings suggest that mental illness is associated with dysregulation of immune, metabolic, and fundamental cellular regulatory processes in the peripheral blood (Supplementary Figure S3).

We then examined transcriptional differences between SSD and BP subjects, both on and off APs, and healthy controls. The analysis identified 2,154 and 1,611 genes that were differentially expressed in subjects off and on APs, respectively, compared to HC (Figures 1B,C) (Supplementary Tables S3-S4). We then compared subjects on versus off APs; this analysis identified only four genes that were differentially expressed, with one upregulated gene, CYRIA, and three downregulated genes, TSIX, FRS2, and XIST (Supplementary Table S5).

Venn diagram analysis revealed 1,189 common upregulated genes and 89 downregulated genes between the MD and MD + APs DEGs (Supplementary Figure S4A). Functional enrichment analysis was conducted using Metascape and represented in the heatmap (Figure 2A). We identified significant enrichment of immune- and infection-related pathways, such as antigen receptor-mediated signaling, regulation of pattern recognition receptor signaling, and lymphocyte differentiation, as well as processes related to RNA metabolism and protein regulation, including protein stabilization and postsynaptic density organization. Additionally, we observed enrichment in signaling and lipid metabolic pathways, including sphingolipid signaling, phosphatidylinositol phosphate synthesis, and metabolic processes such as the one-carbon folate pool, DNA metabolic process, and the PID insulin pathway. These findings suggest that both the MD and MD + APs groups share a core set of dysregulated pathways involving immune activation, RNA and protein homeostasis, and metabolic regulation. Disease enrichment analysis also revealed strong links to immune-related, infectious, and neoplastic disorders. The gene sets were associated with conditions such as cutaneous melanoma and female breast carcinoma and showed connections to neurological and metabolic conditions, including encephalitis and progressive encephalopathy (Figure 2B).

The enrichment analysis of protein-protein interactions (PPI) among the common DEGs is shown in Supplementary Figure S4B. The MCODE plugin, a tool for identifying functional modules in PPI networks, was used in this analysis. The top-scoring modules were “RHOD GTPase cycle,” “generation of second messenger molecules,” and “cytokine signaling in the immune system” (Supplementary Table S6).

To understand hidden patterns in the expression data and identify potential subgroups of genes with collective classification value, we performed WGCNA. We identified the module with the greatest differential expression across the groups (MD and MD + APs). This analysis revealed that module M1 has elevated expression in mentally ill subjects compared to healthy controls (P_adj = 8.017x10⁻¹⁸) (Supplementary Figure S5). Next, we checked whether the genes in M1 overlap with DEGs in MD and MD + APs compared with HC (Supplementary Table S7). We observed 966 common genes across the three datasets. These genes were enriched in immune-related pathways, such as lipopolysaccharide-mediated signaling and antigen receptor regulation, as well as metabolic and genomic processes, including folate metabolism, telomere maintenance, and ATR activation in response to replication stress. Disease enrichment further linked these genes to spinocerebellar ataxia, classical Hodgkin’s lymphoma, and acute coronary syndrome, indicating their relevance to both neuropsychiatric pathology and systemic comorbidities (Supplementary Figure S6).

Our sample showed no significant differences in gender or age (Table 1). The smoking history showed a significant difference (Χ² = 12.5, p = 0.002), and post hoc comparisons revealed that the proportion of smokers was significantly lower in the HC than in the MD + AP group (p < 0.05) (Table 1). The duration of mental illness was significantly different in the two patient groups, but the onset of symptoms and diagnosis was significantly earlier in the MD + APs than the MS off APs (p < 0.05) (Table 1).

We compared a panel of metabolic clinical features among MD, MD + APs, and HC. BMI, waist circumference, fasting blood glucose, HbA1c, triglycerides, and HDL cholesterol did not differ significantly across the groups (Table 2; Supplementary Figure S7). In contrast, we observed significant group effects for LDL cholesterol (ANOVA, p = 0.001), systolic blood pressure (ANOVA, p = 0.007), and diastolic blood pressure (ANOVA, p = 0.004). Post-hoc comparisons (Table 2) revealed that LDL cholesterol levels were higher in MD + APs than in MD, while diastolic blood pressure was elevated in both MD and MD + APs compared to HC. Systolic blood pressure was only increased in MD + APs compared to HC (Figure 3). These findings suggest that cardiovascular risk markers, especially lipids and blood pressure, were altered in patients with or without antipsychotic treatment. We used WGCNA to explore the relationships between identified modules and clinical metabolic features. The module–trait heatmap revealed specific associations: ME126 was strongly correlated with waist circumference (r = 0.97, q = 1.4 × 10⁻⁹³), while ME12 correlated with BMI (r = 0.36, q = 3.6 × 10⁻⁴), indicating gene clusters potentially linked to obesity-related pathways. Additionally, ME14 and ME71 both associated with HbA1c (r = 0.36, q = 2.4 × 10⁻⁴; r = 0.36, q = 3.5 × 10⁻⁴). ME32 correlated with fasting glucose (r = 0.46, q = 1.9 × 10⁻⁷), and ME122 was associated with triglyceride levels (r = 0.31, q = 5.2 × 10⁻³). HDL cholesterol showed correlations with ME138 (r = 0.51, q = 2.2 × 10⁻⁹) and ME89 (r = 0.33, q = 1.5 × 10⁻³). Blood pressure and pulse also demonstrated significant associations, with ME109 linked to systolic (r = 0.39, q = 4.9 × 10⁻⁵) and diastolic blood pressure (r = 0.38, q = 8.8 × 10⁻⁵). Overall, these results highlight co-expressed genes connected to metabolic and cardiovascular traits in our cohort (Supplementary Figure S8; Supplementary Table S8).

We then subcategorized the subjects with mental illness based on their diagnosis and compared the different groups. Comparing BP and BP + APs versus HC yielded 2,229 and 1936 DEGs, respectively (Supplementary Figure S9; Supplementary Tables S9-S10). A Venn diagram of the significant DEGs in both BP and BP + APs showed 1,273 shared upregulated genes and 100 downregulated genes (Supplementary Figure S10A). We performed functional enrichment analysis using Metascape, which is represented in the heatmaps (Figures 4A,B). This analysis revealed clusters of genes involved in immune-related and metabolic pathways. Notable terms included lymphocyte differentiation, regulation of antigen receptor-mediated signaling, negative regulation of the secretory pathway, sphingolipid signaling, and DNA repair. Disease enrichment analysis of the shared gene set indicated associations with various immune-mediated and systemic disorders, such as autoimmune hepatitis, primary biliary cirrhosis, systemic lupus erythematosus, connective tissue diseases, and HIV-1 infection. Additionally, enrichment in progressive encephalopathy, retinitis, and cutaneous melanoma suggests links to neurodegenerative and cancer-related pathways. MCODE clustering identified 11 functional modules, with key enrichments including small GTPase cycles and mitotic processes (MCODE1), purine and nucleotide metabolism (MCODE2), sphingolipid and ceramide metabolism (MCODE3), and snRNA processing (MCODE4) (Supplementary Figure S10B; Supplementary Table S11).

We then conducted differential expression analysis in SSD patients, both on and off APs, compared to HC, identifying 685 and 1,047 DEGs, respectively (Supplementary Tables S12, S13). A Venn diagram analysis showed 484 common upregulated and only seven common downregulated genes. Due to the small number of downregulated genes, enrichment analysis was performed only on the upregulated set. Functional annotation revealed significant enrichment in pathways related to hexose catabolic processes, telomere maintenance in response to DNA damage, folate metabolism and transport, rRNA modification, and pyruvate metabolism. Disease association analysis indicated enrichment in conditions such as benign prostatic hyperplasia, spinocerebellar ataxia, and pancreatic ductal carcinoma. These findings suggest that the transcriptomic changes common to SSD, independent of antipsychotic exposure, involve cellular metabolic pathways, genome maintenance, and immune-oncogenic processes (Supplementary Figure S11).

To assess the reproducibility of our results, we compared the differentially expressed genes (DEGs) identified in SSD patients (both on and off APs) with those reported in three transcriptomic studies (Pan et al., 2025; Wang et al., 2024; Cui et al., 2021). The overlap analysis showed that six genes (MATN1, PLIN4, DNAJB1, SLC22A4, CRYZ, and LIM2) were shared between our dataset and Pan et al., study (Pan et al., 2025), while another six genes (MFSD3, CSNK1D, MIB2, TMEM238, GPR152, and HRNR) overlapped with Wang et al., study (Wang et al., 2024). No genes overlapped with Cui et al., study (Cui et al., 2021). Notably, one gene, SPON2, was common across all our datasets and both studies by Pan et al. and Wang et al. (Pan et al., 2025; Wang et al., 2024). These overlapping genes may be robust candidates underlying the molecular pathology of SSD, irrespective of treatment status. We then compared the 1,376 DEGs common in BP patients on and off APs with transcriptomic signatures from three other studies. Only one gene, OGT, was shared between our dataset and Hess et al. study (Hess et al., 2020), with no overlapping genes observed with Krebs et al. or Munkholm et al. studies (Krebs et al., 2020; Munkholm et al., 2015). Each of these published studies showed different gene sets, highlighting the heterogeneity of blood-based transcriptomic profiles (Krebs et al., 2020; Munkholm et al., 2015) (Supplementary Figure S12).