This study utilized a biobank established in our previous research (Registration No.: ChiCTR2100049660), which was approved by the Ethics Committee of the Affiliated Hospital of Shandong University of Traditional Chinese Medicine [(2021) Ethics Approval No. (055)-KY]. All procedures involving human participants were conducted in accordance with institutional ethical guidelines, and written informed consent was obtained from all participants prior to enrollment.

Participants were recruited from the Affiliated Hospital of Shandong University of Traditional Chinese Medicine, including 64 individuals with SD and 32 healthy controls (HC). Initial screening was performed using the Patient Health Questionnaire-9 (PHQ-9), followed by the Mini International Neuropsychiatric Interview (MINI, version 5.0) to confirm depressive symptoms while ensuring that SD participants did not meet the DSM-V criteria for MDD. Inclusion criteria for the SD group were as follows: age 18–60 years; Hamilton Depression Rating Scale (HAMD-17) scores between 7 and 16; and no prior treatment for depression or use of antibiotics, probiotics, or fermented dairy products within the preceding two weeks. Exclusion criteria were diagnoses of MDD, bipolar disorder, or psychotic disorders; severe hepatic or renal dysfunction; serious arrhythmia or cardiac insufficiency; pregnancy; depression secondary to psychoactive substances or neurological diseases; suicidal ideation or attempts; or inability to comply with study procedures. HC participants were required to provide a recent medical check-up report and to be free of psychiatric disorders according to the MINI interview, and they were subjected to the same exclusion criteria as the SD group. Demographic and clinical information—including age, gender, body mass index (BMI), emotional status, exercise habits, and alcohol/tobacco use—was collected from all participants. Chi-square and Fisher’s exact tests revealed no significant between-group differences in gender distribution, age, BMI, emotional status, exercise frequency, or smoking behavior (Supplementary Table 1).

Fresh stool samples (~1 g) were collected using sterile sampling spoons and transferred into pre-labeled 2-ml cryovials (Biologix, USA), which were immediately stored in liquid nitrogen until analysis. Participants fasted for at least 8 hours prior to blood sampling, and female participants avoided sample collection during menstruation. On the following morning (8:00–9:00 AM), approximately 4 ml of fasting venous blood was drawn into EDTA anticoagulant tubes. Samples were centrifuged at 3000 rpm for 10 minutes at 4 °C, and plasma was aliquoted into cryovials and stored at −80 °C for subsequent metabolomic analysis.

Microbial DNA was extracted from stool samples using the E.Z.N.A.^® Soil DNA Kit (Omega Bio-tek, USA). The full-length 16S rRNA gene was amplified with primers 27F and 1492R (18) and sequenced on the PacBio Sequel II platform (Pacific Biosciences, USA) at Majorbio (Shanghai, China). CCS reads were processed with the DADA2-CCS plugin (19) in QIIME2 (20). Samples were rarefied to 6,000 reads, achieving an average Good’s coverage of 99.09%. Amplicon sequence variants (ASVs) were assigned using the SILVA v138 database. Alpha diversity metrics, including ACE, Chao, Shannon, Simpson, Shannon evenness, and Simpson evenness indices, were calculated to evaluate microbial diversity, and inter-group differences were analyzed using the Wilcoxon test. Beta diversity was assessed using Principal Coordinates Analysis (PCoA) with weighted and unweighted UniFrac distances, and group differences were tested with the Adonis (PERMANOVA) test. Spearman correlations (|r| > 0.2, P < 0.05) were used to identify genera for network analysis (21). Statistical analyses were performed in R 4.3.1 and the Majorbio Cloud Platform (22).

Plasma metabolites were extracted using a precooled extraction solution and processed by vortexing, ultrasonic-assisted extraction, and centrifugation. The resulting supernatants were analyzed using ultra-high-performance liquid chromatography–mass spectrometry (UHPLC–MS). Raw data were processed with Progenesis QI (Waters Corporation, USA) and matched against HMDB, Metlin, and the Majorbio in-house database for metabolite identification. Features detected in <80% of samples were removed, and peak intensities were normalized. Variables with a relative standard deviation (RSD) >30% in quality control samples were excluded, and data were log10-transformed before statistical analysis. Multivariate analysis was conducted using OPLS–DA in the ropls R package (version 1.6.2), complemented by Wilcoxon rank-sum tests and fold-change analysis. Metabolites were considered significantly altered when meeting both criteria: VIP >1 and p < 0.05. Differential metabolites were mapped to biochemical pathways through KEGG-based enrichment and pathway topology analyses, and categorized according to pathway and functional classes.

The identified endogenous differential metabolites were treated as environmental variables and correlated with the top 30 bacterial genera ranked by relative abundance. Associations between bacterial genera and metabolic features were evaluated using Spearman correlation coefficients.

To characterize the gut microbiota features of individuals with SD, full-length 16S rRNA sequencing was performed on 96 fecal samples. After sequence denoising, a total of 11,184 ASVs and 1,804,258 high-quality reads were obtained, with an average of 18,794 reads per sample. The core ASV accumulation curve reached a plateau (Supplementary Figure S1A), indicating that the sample size was adequate. Rarefaction curves approached saturation (Supplementary Figures S1B, C), suggesting sufficient sequencing depth. At the ASV level, no significant differences were observed in richness indices, including Chao1 and ACE (Supplementary Figures S1D, E). However, the SD group exhibited significantly lower Shannon and Shannon evenness indices compared with the HC group (P < 0.05 and P < 0.01, respectively; Figures 1A, B), reflecting reduced gut microbial diversity and community evenness. Principal coordinate analysis (PCoA) based on weighted and unweighted UniFrac distances further revealed significant separation between the SD and HC groups (both P < 0.01; Figures 1C, D). Similar clustering patterns were observed at the genus level (Supplementary Figures S1F. G). Correlation network and module analyses showed that the HC microbiota network tended to be scale-free with fewer inter-module connections, whereas the SD network was more complex, characterized by increased connectivity within and between modules (Figures 1E, F). These results collectively demonstrate pronounced alterations in gut microbiota composition and structure in individuals with SD compared with HC.

Further differential abundance analysis revealed marked alterations in the gut microbiota between the SD and HC groups. Compared with the HC group, the SD group showed significant decreases in Firmicutes and Actinobacteriota, together with significant increases in Proteobacteria and Synergistota (Figure 2A). At the genus level, the SD group exhibited significantly lower abundances of Blautia, Eubacterium hallii group, Dorea, Agathobacter, Anaerostipes, and Ruminococcus gauvreauii group, whereas Escherichia–Shigella, Monoglobus, Lachnoclostridium, and Enterobacter were significantly enriched (Figure 2B). These results were further supported by LEfSe analysis (Supplementary Figure S1H). Taken together, these findings indicate that although depressive symptoms in SD individuals are relatively mild, distinct alterations in gut microbiota composition are already evident, suggesting a potential involvement of gut dysbiosis in the early stages of depression.

OPLS-DA analysis demonstrated clear separation between the plasma metabolic profiles of individuals with SD and HC (Figures 3A, B), and the robustness of the model was supported by OPLS-DA permutation testing (Supplementary Figures S2A, B). These results indicate substantial differences in plasma metabolic phenotypes between the two groups. Based on P < 0.05 and VIP ≥ 1, a total of 615 differential metabolites were identified.

To explore the biological significance of these SD-related metabolites, we performed KEGG annotation and obtained 112 annotated plasma metabolites (Supplementary Table S2). KEGG compound classification showed that 49 metabolites mapped to eight biologically relevant categories, with lipids being the most abundant, followed by organic acids (Figure 3C). Furthermore, 54 compounds were assigned to six lipid subclasses, with glycerophospholipids being the most prevalent, followed by fatty acyls, sterol lipids, sphingolipids, prenol lipids, and glycerolipids (Figure 3D). KEGG pathway topology analysis revealed enrichment of several key metabolic pathways, including linoleic acid metabolism, glycerophospholipid metabolism, caffeine metabolism, biotin metabolism, sphingolipid metabolism, nucleotide metabolism, the citrate (TCA) cycle, pyrimidine metabolism, lysine degradation, and phenylalanine metabolism (Figure 3E, Supplementary Table S3). Eight pathways exhibited significant differences: biotin metabolism, nucleotide metabolism, lysine degradation, phenylalanine metabolism, and pyrimidine metabolism were elevated overall, while linoleic acid metabolism, glycerophospholipid metabolism, and sphingolipid metabolism showed overall decreases (Figure 3F).

To assess the diagnostic potential of the identified metabolites, ROC curve analysis was performed. Eight endogenous metabolites had an AUC greater than 0.9 (Supplementary Figures S2 C–I, Supplementary Table S4), including three involved in lipid metabolism. The combined ROC curve for these metabolites produced an AUC of 0.993, with 96% sensitivity and 95% specificity (Figure 3G), indicating that this metabolite panel may serve as a promising biomarker set for the clinical identification of SD.

The most important differential metabolites were first identified using random forest analysis (Supplementary Table S5), followed by Spearman correlation analysis to assess associations between plasma metabolites and gut microbiota. The results showed that Eubacterium hallii group, Blautia, Dorea, and Agathobacter, which were reduced in abundance in SD subjects, were significantly positively correlated with decreased plasma levels of N-Formyl-L-glutamic acid, PE[18:0/20:1(11Z)], Aldosterone, and PC[15:0/22:1(13Z)], and negatively correlated with several upregulated plasma metabolites. In healthy individuals, Lachnospiraceae and Escherichia–Shigella, which were also lower in abundance, displayed the opposite correlation pattern with plasma metabolites (Figure 4).

Androst-5-ene-3β,17β-diol, 5-Androstenediol, and Aldosterone are metabolites involved in steroid hormone biosynthesis. PE[16:1(9Z)/P-18:1(11Z)], PE[18:0/20:1(11Z)], PG[i-12:0/18:2(9Z,11Z)], and PC(15:0/22:1(13Z)) participate in glycerophospholipid metabolism, whereas Phytosphingosine and 4-hydroxysphinganine are components of sphingolipid metabolism. These findings suggest coordinated alterations between gut microbiota and host metabolic pathways in SD, particularly those related to lipid metabolism.