This study received approval from the Ethics Committee of the Xianyang Hospital of Yan’an University (YDXY-KY-2020–016). Fasting serum samples were acquired from depression patients at the Xianyang Hospital of Yan’an University’s Department of Psychiatry from September 2020 to September 2022. Likewise, fasting serum samples from healthy controls were sourced from the Department of Physical Examination at Xianyang Hospital of Yan’an University during the same period. Informed consent was obtained from the patients and volunteers. The criteria for depression included patients aged 8 to 80 years who had a diagnosis of depression based on the Chinese Classification and Diagnostic Criteria of Mental Disorders (CCMD), the criteria and diagnosis of depression in the CCMD are reliable and agree well with ICD-10. In this study, the depressive patients included were all primary depressive patients who had not received any treatment before the onset of the illness and had no other relevant psychiatric comorbidities to the onset. Patients with tumors, hematological system diseases, severe immune system disease or heart, kidney, liver serious insufficiency, congenital, or hereditary diseases were excluded.

A total of 192 serum samples were collected from 96 depression patients, median (IQR) age 57(39.75, 65), 31.25% male, 68.75% female, and 96 healthy controls, median (IQR) age 59(46, 65.25), 35.42% male, 64.58% female. According to the 1:1 ratio, the 192 samples were randomly divided into the training group (48 depression patients, 48 healthy controls) and the verification group (48 depression patients, 48 healthy controls). The training group was used to screen the markers by mass spectrometry, and the validation group was used to validate the diagnostic efficacy of the candidate markers by ELISA. The baseline characteristics of the depression patients and healthy controls are shown in Tables 1 , 2 and Supplementary Table 1 .

All serum samples were collected in 5 mL vacuum tubes without anticoagulant, then the samples were allowed to stand at 4 °C for 1 h, centrifuged at 3000 xg for 20 min at 4 °C and stored at -80 °C until use.

Serum samples were separated using MB-WCX (ClinProt purification reagent kits; Bruker Daltonics, Bremen, Germany). According to the standard protocol described by Bruker, the sample was bound to the magnetic beads. The magnetic beads were thoroughly mixed on a vortexing device for 1 minute.10 µL of binding buffer and 10 µL of MB-WCX beads were transferred to a standard thin-walled PCR tube and mixed by pipetting up and down.5 µL of serum was added to the solution and mixed intensively by pipetting up and down five times. The tube was left to incubate for 5 minutes. The tube was placed in the magnetic separator and the beads were collected on the wall of the tube for 1 minute. Carefully remove the supernatant with a pipette. Avoid contact of the pipette tips with the beads and take care not to remove the beads. Add 100 µL Wash Buffer to the tube. Collect the beads at the tube wall for 1 minute. Carefully remove the supernatant with a pipette. Repeat washing of the magnetic beads twice. Add 5 µL elution buffer and dissolve the beads from the tube wall by vigorous pipetting up and down 10 times. Collect the beads at the tube wall for 2 minutes. Transfer the clear supernatant to a fresh tube. Add 5 µL of stabilization buffer to the eluate and mix vigorously by pipetting up and down. The sample to be tested is obtained. Then apply 1 µL of sample to a target spot and allow to dry at room temperature (3–10 min). Apply 1 µL of matrix consisting of α-Cyano-4-hydroxycinnamic acid (HCCA) (3 mg/mL, in 50% acetonitrile/2% TFA) and allow to dry at room temperature. Each sample was spotted in triplicate to evaluate the reproducibility of the method.

Targets were immediately analyzed on a calibrated Autoflex III MALDI-TOF-MS (Bruker) using flexControl version 3.0 software (Bruker) and an optimized measurement protocol. Mass calibration was performed using a standard calibration mixture of peptides and proteins (mass range: 0.8–10 kDa). All assays were performed in a blinded fashion, including serum analyzes of different groups. Data analysis was performed using FlexAnalysis software (version 3.0; Bruker). Peptide pattern recognition was performed using ClinProTools version 2.2 (Bruker Daltonics), including spectral pretreatment, peak selection and peak calculation operations.

The peptides eluted from the magnetic beads and were analyzed by nano-UPLC-ESI-MS/MS using a nano Aquity UPLC (Waters Corporation, Milford, USA) coupled to a Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). Samples of 20μL (the sample was diluted by 2 times) were loaded on a C18 precolumn (Symmetry®C18, 5 μm, 180 μm × 20 mm, nanoAcquity™Column) at 15 μL/min in 5% acetonitrile (Sigma-Aldrich, St Louis, MO, USA), 0.05% trifluoroacetic acid (Sigma-Aldrich) for 3 min. The precolumn was switched online with the analytical column (Symmetry®C18, 3.5 μm, 75 μm × 150 mm, nanoAcquity™ Column) equilibrated in 95% solvent A (5% acetonitrile, 0.1% formic acid; Sigma-Aldrich) and 5% solvent B (95% acetonitrile, 1.2% formic acid). Peptides were eluted using a 5% to 80% gradient of solvent B over 60 min at a flow rate of 600 nL/min. The Orbitrap Fusion mass spectrometer was operated in the data-dependent mode to switch automatically between MS and MS/MS acquisition. Full-scan survey MS spectra with 2 microscans (m/z 200–2000) were acquired with the Orbitrap with a mass resolution of 100000 at m/z 400, followed by 10 sequential LC-MS/MS scans. Dynamic exclusion was used with 2 repeat counts, 10 s repeat duration and 60 s exclusion duration. For MS/MS, charge state 1 was rejected and precursor ions were activated using 25% normalized collision energy at the default activation q of 0.25. The mass spectra were searched against the human Uniprot database (https://www.uniprot.org/) using Proteome Discoverer software (Version 2.5.0); To reduce false positive identification results, a decoy database containing the reverse sequences was appended to the database. The parameters for the search were as follows: no enzyme, the variable modification was oxidation of methionine, peptide tolerance, 20 ppm, MS/MS tolerance, 1.0 Da.

We used gene ontology (GO) enrichment analysis to analyze the identified proteins. We used the STRING database as a source to build an interaction network of potential biomarkers.

All serum sample assays were blinded, and the standards for the ELISA kits and samples were run in triplicate. The concentrations of TUBB, ITIH4, C3 and C4A were quantified using the human TUBB ELISA kit (No. HB2558-Hu, Shanghai Hengyuan Biotech, China), the human ITIH4 ELISA kit (No. HB2554-Hu, Shanghai Hengyuan Biotech, China), the human C3 ELISA Kit (No. HB1872-Hu, Shanghai Hengyuan Biotech, China), the human C4A ELISA Kit (No. HB1508-Hu, Shanghai Hengyuan Biotech, China), according to the manufacturer’s instructions. The OD was measured at 450 nm.

Statistical analysis was performed using SPSS version 27.0 and GraphPad Prism version 8.0 (GraphPad Software, San Diego, CA, USA), with all data presented as mean ± standard deviation or median and interquartile range. Statistical significance was set at p < 0.05. t-tests, one-way ANOVA, Mann-Whitney U tests and Kruskal-Wallis H tests were used to compare quantitative data. Chi-square tests were used to compare categorical variables. Based on the univariate analysis, statistically significant variables were selected for logistic regression analysis. Predictive variables were identified and an ROC curve was plotted to determine the AUC value.

To ensure reproducibility and stability of the mass spectra, we analyzed all samples in triplicate and found highly reproducible peaks ( Figures 2A, B ). MB-WCX and MALDI-TOF-MS were used to compare the proteomic profiles of depression patients and healthy controls. Fractionation of serum samples showed that depression patients (red) and healthy controls (green) had proteomic profiles from 0.8 to 10 kDa ( Figures 2A, B ). The differentially expressed peaks between the two groups were detected in this mass range. According to the serum proteomics analysis, only a few overlapping areas were found between depression patients and healthy controls, indicating that the groups could be accurately distinguished ( Figures 2C, D ).

A total of 96 peaks, expressed as mean ± standard deviation, were identified in all groups analyzed. The information about all 96 peaks was added in Supplementary Table 2 . The expressing levels of 96 peaks were showed by a heat map ( Supplementary Figure 1 ). After obtaining 96 peaks, we further screened them based on fold change (FC) and significance level (P-value). The threshold standard was set as FC > 1.5 or FC < 0.67 and p < 0.05 (|log₂FC| > 0.58 & p < 0.05). Subsequently, a volcano plot was generated, revealing that the depression group exhibited significant up-regulation of 5 peaks and down-regulation of 16 peaks( Supplementary Figure 2 ). Considering the significant time and financial costs associated with annotating all 21 candidate peptides that are significantly up- or down-regulated using LC-ESI-MS/MS, as well as the challenges of detecting low-abundance down-regulated peptides with low sensitivity and susceptibility to technical variability and measurement noise leading to false negative results, which can compromise diagnostic accuracy in clinical serum biomarker detection, we have selected a subset of upregulated peptides for further molecular identification annotation. Specifically, five peaks (peak 1, m/z: 1868.21; peak 2, m/z: 1062.35; peak 3, m/z: 1452.12; peak 4, m/z: 1208.72; peak 5, m/z:1619.58) were chosen for further analysis ( Figures 3A–E ).

All five peptide peaks (peak 1, m/z: 1868.21; peak 2, m/z: 1062.35; peak 3, m/z: 1452.12; peak 4, m/z: 1208.72; peak 5, m/z: 1619.58) that differed significantly between healthy controls and depression patients ( Table 3 ) were identified using LC-ESI-MS/MS and the UniProt database. We identified proteins from the MS/MS spectra of these peptides, including TUBB, ITIH4, C3, C4A ( Table 4 ).

The mean serum concentrations of ITIH4, C3, C4A, TUBB were 154.77 ± 100.75 pg/mL, 100.70 ± 32.89 μg/mL, 283.20 ± 97.17 μg/mL, 129.78 ± 72. 80 ng/mL in depression patients; 52.11 ± 32.23 pg/mL, 65.03 ± 21.40 μg/mL, 171.50 ± 55.37μg/mL, 66.17 ± 30.48 ng/mL in healthy controls; ( Figures 4A–D ). The median serum concentrations of ITIH4 were 113.22 pg/mL, 46.57 pg/mL in depression patients and healthy controls, respectively, with a significant difference between the two groups ( Figure 4A ). The median serum concentrations of C3 were 93.03μg/mL, 64.77μg/mL in depression patients, healthy controls, respectively, with significant difference between the two groups ( Figure 4B ). The median serum concentrations of C4A were 269.05μg/mL, 165.11μg/mL in depression patients, healthy controls, respectively, with significant difference between the two groups ( Figure 4C ). The median serum concentrations of TUBB were 110.42 ng/mL and 69.20 ng/mL in depression patients and healthy controls, respectively, with a significant difference between the two groups ( Figure 4D ).

Univariate analysis showed that ITIH4, C3, C4A and TUBB levels were significantly different between depression patients and healthy controls. ROC curves of ITIH4, C3, C4A and TUBB are shown in Figures 5A–D . Using serum levels of ITIH4, C3, C4A and TUBB, the AUC of the ITIH4 ROC curve was 0.924 (95% CI, 0.870–0. 978; p < 0.001, sensitivity 91.67%; specificity 83.33%; cutoff = 68.33; Figure 5A ), the AUC of the C3 ROC curve was 0.826(95% CI, 0.743–0.909; p < 0.001, sensitivity 89. 58%; specificity 66.67%; cut-off value = 82.50; Figure 5B ), the AUC of the C4A ROC curve was 0.853(95% CI, 0.780–0.926; p < 0.001, sensitivity 72.92%; specificity 83. 33%; cut-off value = 195.98; Figure 5C ), the AUC of the TUBB ROC curve was 0.825 (95% CI, 0.745–0.905; p < 0.001, sensitivity 89.58%; specificity 58.33%; cut-off value = 99.50; Figure 5D ). Multivariate logistic regression analysis identified serum ITIH4, C3, C4A and TUBB levels as independent risk factors for depression. Using the combination of serum ITIH4, C3, C4A and TUBB levels, the AUC of the ROC curve was 0.973 (95% CI, 0.946–0.999; p < 0.001, sensitivity 95.83%; specificity 91.67%; cut-off value = -0.536; Figure 5E ).

Pearson’s analysis was used to evaluate the correlation between ITIH4, C3, C4A, TUBB, AST, ALP, GGT, TBA, ALT, TP, ALB, GLO, TBIL, ADA, GLDH, CHE; TUBB levels were significantly correlated with blood ITIH4 levels (r =0.435, p = 0.002) and blood GGT levels (r =0.431, p = 0.002). C3 levels correlated significantly with blood ADA levels (r = 0.320, p = 0.027).

The GO database analyzes gene classification annotation and biological function by biological process (BP), cellular component (CC) and molecular function (MF). The GO annotations for the four proteins were analyzed, and the results are shown in Figure 6 . The four proteins were mainly involved in biological processes such as regulation of apoptotic cell clearance, negative regulation of peptidase activity, negative regulation of endopeptidase activity, lymphocyte-mediated immunity, negative regulation of proteolysis. The four proteins were mainly existed in cellular component such as blood microparticle, vesicle lumen, cytoplasmic vesicle lumen, secretory granule lumen, azurophil granule lumen. The four proteins were mainly involved in molecular function such as endopeptidase regulator activity, peptidase inhibitor activity, endopeptidase inhibitor activity, peptidase regulator activity, enzyme inhibitor activity.

To investigate the alteration of the protein interaction network in depression, we extracted networks using STRING software. A protein-protein interaction network was constructed using four seed proteins and their ten predicted functional partners in Figure 7 . C-C chemokine receptor type 5 (CCR5), Cluster of differentiation 4 (CD4), Membrane cofactor protein (CD46), Complement decay-accelerating factor (CD55), Complement factor H (CFH), Complement factor H-related protein 1 (CFHR1); Properdin (CFP), Complement receptor type 1 (CR1), Complement receptor type 2 (CR2), Detyrosinated tubulin alpha-1B chain (TUBA1B).