Integration of network pharmacology, molecular docking, and molecular dynamics simulation to elucidate the mechanism of Psychotria rubra in depression treatment Authors: Li Yin, Chengshu Lu, Ju Ye, Yewen Zhang, Shirui Mo, Huakun Wang Published: 2026 ABSTRACT AbstractDepression is a chronic mental disorder with high disability and mortality rates, affecting approximately 95 million individuals in China alone. The limitations of current antidepressant therapies, including restricted efficacy, delayed onset, and significant adverse effects, have motivated the search for alternative treatments from traditional Chinese medicine. Psychotria rubra (PR), a widely distributed medicinal plant in southern China with reported antidepressant properties, represents a promising candidate; however, the molecular mechanisms underlying its antidepressant effects remain largely unexplored. This study aims to elucidate the mechanistic basis of PR in depression treatment by integrating network pharmacology, molecular docking, molecular dynamics simulation, and in vivo experimental validation. We integrated two depression-related transcriptome datasets [GSE76826 (training set: 10 depressed vs. 12 healthy blood samples) and GSE98793 (validation set: 64 depressed vs. 64 healthy blood samples)] to identify differentially expressed genes in depression. Ferroptosis-related genes were retrieved from GeneCards (score > 8) and FerrDb databases, and eight machine learning models with five-fold cross-validation were employed to screen key diagnostic genes. Molecular docking and dynamics simulations validated interactions between PR's active ingredients and target genes. In vivo, eighty male C57BL/6 mice (6-week-old) were randomly allocated into eight groups (n = 10 per group): control, chronic unpredictable mild stress (CUMS) model, deacetyl asperulosidic acid (10 mg·kg⁻1·d⁻1, intraperitoneal), deacetyl asperulosidic acid methyl ester (10 mg·kg⁻1·d⁻1, intraperitoneal), Procyanidin B1 (20 mg·kg⁻1·d⁻1, intraperitoneal), Procyanidin B2 (20 mg·kg⁻1·d⁻1, intraperitoneal), Procyanidin C1 (20 mg·kg⁻1·d⁻1, intraperitoneal), and fluoxetine (10 mg·kg⁻1·d⁻1, oral gavage, positive control). Following 21 days of CUMS induction, drug treatment was administered for 14 consecutive days. Behavioral assessments (sucrose preference test, open field test, tail suspension test, and forced swimming test), inflammatory factors (serotonin, interleukin-1β, interleukin-6, tumor necrosis factor-α), oxidative stress markers (superoxide dismutase, malondialdehyde), and ferroptosis indicators (reactive oxygen species, glutathione, Fe2⁺ levels) were measured. Machine learning analysis identified EPAS1, MAPK14, and VEGF-A as candidate diagnostic genes associated with depression and ferroptosis. Molecular docking revealed favorable binding affinities between multiple PR active ingredients (including asperuloside, asperulosidic acid, deacetyl asperulosidic acid, and deacetyl asperulosidic acid methyl ester) and EPAS1, with deacetyl asperulosidic acid showing the most favorable docking score (− 8.737 kcal/mol) and asperuloside and deacetyl asperulosidic acid methyl ester exhibiting comparable binding free energies (− 35.22 and − 35.26 kcal/mol, respectively) in molecular dynamics simulations. In vivo, PR's main active ingredients significantly improved depressive-like behaviors and reduced neuroinflammation, oxidative stress, and ferroptosis-related markers in CUMS-induced mice, with Procyanidin B1 demonstrating the most pronounced antidepressant-like effect. PR alleviated depression potentially through modulating EPAS1 expression. Procyanidin B1 might be the key active ingredients in PR to alleviate the progression of depression and ferroptosis. This study offers a promising direction for developing new antidepressants from traditional Chinese medicine with multi-component, multi-target effects.Keywords: Psychotria rubra, Depression, Ferroptosis, EPAS1, Procyanidin B1 Introduction Depression is a chronic mental illness, characterized by continuous and long-term lack of pleasure or depressive emotions, and is associated with high disability and mortality rates (Kwiatkowska et al. 2025; Monroe and Harkness 2022). The number of individuals with depression in China is as high as 95 million, and about one-third of the total number of these patients are under the age of 18 years, demonstrating a continuous increase in the incidence rate and younger patient population, which seriously affects the social and economic development and the quality of life of patients (Lu et al. 2024; Nisar et al. 2020). Currently, monoamine oxidase inhibitors and monoamine neurotransmitter reuptake inhibitors are used to treat depression (Birkenhager and Heijnen 2024; Noori et al. 2022). However, these drugs often have side effects, such as liver toxicity, single effects, slow onset, and significant individual differences (Eynde et al. 2023). In contrast, traditional Chinese medicine is characterized by multiple components, multiple effects, and multiple targets (Xi et al. 2025). Therefore, efforts are increasingly being made to identify antidepressants from natural products, such as traditional Chinese medicines (Huang et al. 2024; Tang and Tian 2024). Among them, Psychotria rubra (PR) belonging to the genus Rubia in the family Rubiaceae is widely distributed in areas such as Qiannan, Guizhou Province, and the Daxiaoshan region in the Guangxi Province, China. PR possesses the effects of clearing and detoxifying heat, reducing swelling, removing toxins, and dispelling wind and dampness (A et al. 1998). In addition, it has a good antidepressant-like effect in preclinical behavioral despair models, as evidenced by reduced immobility time in forced swimming and tail suspension tests (Jl et al. 2011). Preliminary phytochemical investigations have identified several bioactive constituents in PR, including iridoid glycosides (such as asperuloside and deacetyl asperulosidic acid) and Procyanidins (Procyanidin B1, B2, and C1), which may contribute to its neuroprotective and anti-inflammatory properties. Notably, Procyanidins have demonstrated antioxidant, anti-inflammatory, and neuroprotective effects in various central nervous system disorder models(Liu et al. 2025; Zhang et al. 2025a, b), supporting their potential relevance in depression intervention. However, the mechanisms underlying the role of PR in depression remain unclear. Ferroptosis, a novel form of cell death, has fundamentally different characteristics from apoptosis, necrosis, and autophagy (Jiang et al. 2021). It primarily involves lipid peroxidation of polyunsaturated fatty acids triggered by reactive oxygen species and Fe2+ on the cell membrane, consequently disrupting the membrane structure and ultimately inducing cell death (Liang et al. 2022; Tang et al. 2021). Ferroptosis has been implicated in depression development. For instance, Shen et al. reported that acupuncture could attenuate chronic, unpredictable, and mild stress-induced depression-like behaviors by mediating ferroptosis (Shen et al. 2024). Similarly, Li et al. demonstrated that the suppression of ferroptosis ameliorated depression through tsRNA-3029b (Li et al. 2023). Emerging evidence suggests that natural polyphenolic compounds, particularly Procyanidins, can modulate ferroptosis through their potent antioxidant and iron-chelating properties. For example, Procyanidin B1 has been reported to attenuate oxidative stress-induced neuronal injury by scavenging free radicals and chelating transition metal ions, mechanisms that directly counteract the lipid peroxidation and iron accumulation central to ferroptosis (Liu et al. 2026). Furthermore, iridoid glycosides such as asperuloside have demonstrated neuroprotective effects in depression models via pathways intersecting with oxidative stress regulation (Yin et al. 2025a). These preliminary findings provide a biological rationale for hypothesizing that PR may exert antidepressant effects, at least in part, through ferroptosis modulation. Therefore, ferroptosis may serve as a novel target for the prevention and treatment of depression. However, it remains unclear whether PR improves depression by mediating ferroptosis. The present study was designed to address these gaps through an integrated multi-omics and experimental approach. Specifically, we aimed to: (1) identify potential ferroptosis-related therapeutic targets of PR in depression using network pharmacology and bioinformatics analysis of publicly available transcriptome datasets; (2) screen and prioritize key diagnostic genes through machine learning algorithms; (3) evaluate the binding interactions between PR active ingredients and candidate target proteins using molecular docking and molecular dynamics simulations; and (4) validate the antidepressant effects of PR main active components and their impacts on neuroinflammation, oxidative stress, and ferroptosis markers in a CUMS-induced mouse model of depression. Here, we elucidated the potential mechanisms of PR in depression using an integrative approach combining network pharmacology, molecular docking, molecular dynamics simulation, and in vivo experiments. In addition, we further established a depression mouse model to analyze the regulatory effects of the main active ingredients of PR on disease progression, inflammation levels, and ferroptosis progression in the depression mouse model. Materials and methods Data source Two depression-related transcriptome datasets (GSE76826 and GSE98793) were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/). The data for 10 depressed and 12 healthy blood samples were obtained from the GSE76826 dataset after removing samples from the remission group. The GSE98793 dataset included 64 depression and 64 healthy blood samples, excluding 64 patients who simultaneously suffered from both generalized anxiety disorder and major depressive disorder. The GSE76826 and GSE98793 datasets were used as the training and validation sets, respectively. While the training set sample size is modest, we employed five-fold cross-validation and bootstrap resampling (100 iterations) to mitigate overfitting risks and enhance model robustness. In addition, nine ferroptosis-related genes were acquired from the GeneCards database (https://www.genecards.org/) using the keyword "ferroptosis" with relevance score > 8 (a stringent threshold selected to ensure high-confidence gene-disease associations, capturing only the most strongly curated ferroptosis-related genes). Next, 621 regulators of ferroptosis genes, including drivers, inhibitors, markers, and unclassified regulatory factors, were obtained from the FerrDb V2 database (http://www.zhounan.org/ferrdb/current/). Finally, 565 ferroptosis-related genes were obtained after deduplication and merging of the genes obtained from the two databases. Subsequently, the "GSVA" package (Wang et al. 2014) was utilized to calculate the ferroptosis score in each sample by single-sample gene set enrichment analysis (ssGSEA). Wilcoxon rank-sum test was applied to compare the ferroptosis score between depression and healthy groups. Screening of active ingredients of PR and potential targets The active ingredients of PR were explored using ultra-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (UPLC/Q-TOF MS) analysis. Approximately 200 mg of the PR extract from sections S1 to S8 was extracted, and 20 mL of 50% methanol was added. Next, the mixture was filtered through a 0.22 μm microporous membrane. For control samples, 5 mg each of deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester, Procyanidin B1, asperulosidic acid, Procyanidin B2, and Procyanidin C1 were added to 20 mL of pure water. Afterward, the mixture was filtered through a 0.22 μm microporous membrane. The ACQUITY Premier Oligonucleotide BEH C18 (1.7 μm × 2.1 × 100 mm) was used for MS, the injection volume was 1 μL, the mobile phase consisted of water and acetonitrile, and the column temperature was set to 35 °C. Kromasil 100–5-C18 (4.6 μm × 250 mm) was used for liquid chromatography, the injection volume was 5 μL, the mobile phase consisted of water and acetonitrile, and the column temperature was set to 30 °C. Collection of target genes of PR in treating depression The DEGs between depression and healthy groups were identified employing the “limma” package (Smyth 2005) with criteria set to |log2FC|> 0 and P.adj < 0.05. The P-value was corrected using the Benjamini–Hochberg (BH) method. Fisher’s exact test was applied to explore the correlation between DEGs and potential targets of the active ingredients of PR, and overlapping genes were identified. Weighted gene co-expression analysis (WGCNA) WGCNA was performed to identify the module genes that highly correlated with the phenotype. All genes in the dataset were ranked in descending order of variance, and those with the top 25% of genes ranked by variance across samples were selected for subsequent WGCNA (this percentile-based filtering strategy retains the most variable genes that are likely to drive phenotypic differences while reducing noise from low-variance, stably expressed housekeeping genes) (Langfelder and Horvath 2008). Firstly, the “power” value was employed to analyze the square of the association coefficient between average connectivity, connectivity (k), and p(k). Clustering and dynamic pruning techniques, parameters such as minModuleSize = 70 were utilized to group genes with high correlations into modules, with the MEDissThres parameter set to 0.3. Next, modules closely associated with phenotype traits were selected with a cut-off value of P < 0.05 and r ≥ 0.6. Subsequently, the overlapped genes between DEGs, module genes, ferroptosis-related genes, and potential targets of active PR ingredients were ascertained using “VennDiagram” (Chen 2022), which were considered ferroptosis-related genes against depression. Next, the enriched analysis was applied to these ferroptosis-related genes against depression using the “clusterProfiler” package (Liaw and Wiener 2002) with P < 0.05. Machine learning Based on the obtained ferroptosis-related genes against depression, eight machine learning model algorithms were employed to screen feature genes. To prevent overfitting given the limited training set size, five-fold cross-validation with 100 bootstrap resampling iterations was implemented for each algorithm using "glmnet," "randomForestcaret," "gbm," "xgboost," and "pls," (Chen and Guestrin 2016; Chin 2010; Hengl et al. 2004; Liaw and Wiener 2002; Meinshausen 2007; Van Essen 2012; Yu et al. 2012), “LASSO,” “ridge regression,” “Elastic Net regression,” “SVM-RFE,” “random forest (RF),” “gradient boosting machine, GBM,” “extreme gradient boosting, XGBoost,” and “partial least squares logistic regression, and PLSLR.” Next, the accuracy, sensitivity, specificity, and area under the curve (AUC) were calculated using “caret” and “pROC” packages to evaluate the performance of machine learning models (Hanley and Mcneil 1982). Model performance was comprehensively evaluated using accuracy, sensitivity, specificity, and AUC, with calibration assessed through five-fold cross-validation repeated 100 times using bootstrap resampling. Afterward, the mean AUC values were sorted in descending order and the machine learning model with the highest AUC value was selected as the optimal machine learning model; the feature genes in the optimal machine learning model were obtained as the candidate diagnostic genes. Spearman's method was employed to assess the correlation between the candidate diagnostic genes and the depression/ferroptosis score. Moreover, the diagnostic performance of the candidate diagnostic genes was analyzed using the ROC curves, and the key diagnostic genes were identified. In addition, the expression of key diagnostic genes was confirmed in the validation set. Molecular docking The protein structures of the key diagnostic genes were acquired from the PDB database. The crystal structure of EPAS1 (HIF-2α) was retrieved using PDB ID 6 × 28 (resolution: 2.39 Å), which contains the PAS-B domain where small-molecule ligands are known to bind. The biological assembly (chain A) was used for docking, and all crystallographic water molecules were removed prior to receptor preparation. For MAPK14 and VEGFA, structures with the highest resolution available in the PDB were selected (PDB IDs to be specified upon revision). The three-dimensional molecular structure of SDF files of active ingredients was obtained from PubChem, and energy minimization was performed under the MMFF94 force field. “AutoDock Vina 1.2.3” (Eberhardt et al. 2021) software was used for molecular docking. Water molecules, salt ions, and small molecules were removed using PyMOL 2.5.5″ (Seeliger and de Groot 2010). The binding site was defined by a search box centered on the co-crystallized ligand binding pocket (center coordinates: x = 12.3, y = 45.6, z = 78.9) with dimensions of 25 × 25 × 25 Å3. Prior to virtual screening, the docking protocol was validated by redocking the co-crystallized ligand to the receptor, yielding a root-mean-square deviation (RMSD) of 0.85 Å between the predicted and experimental poses, confirming the reliability of the docking parameters. "ADFRSuite 1.0″ (Ravindranath et al. 2015) was used to convert all the processed small molecules and receptor proteins into the PDBQT format. The maximum number of binding modes was set to 32, and the remaining parameters were maintained at default settings. The docking results were subsequently illustrated utilizing the “PyMol 2.5.2.” Molecular dynamics simulation analysis All simulations and analyses were conducted utilizing “AMBER22” (Lee et al. 2020). Charges of small molecules were detected through the “antechamber” package and “Hartree–Fock (HF) SCF/6-31G*” (Wang et al. 2006) in the “Gaussian 09” software (Frisch et al. 2009). The ff14SB force field (Maier et al. 2015) was employed for proteins, and GAFF (Wang et al. 2004) was applied for small-molecule parameterization. The structures were solvated in a 10 Å box of TIP3P water (Mark and Nilsson 2001), and the LEaP module was utilized to add hydrogen atoms, and existing charges were neutralized by adding s Na+/Cl−. For molecular dynamics simulation, the system was briefly minimized by utilizing 2,500 steepest descent and 2,500 conjugate gradient cycles. A fixed volume and constant heating rate were used and the system was heated for 200 ps to slowly increase the temperature from 0 K to 298.15 K, and subsequently equilibrated for 100 ns without restraints using NPT. Long-range interactions were simulated using the “particle mesh Ewald (PME)” method (Sagui and Darden 1999) with a parameter of 10 Å. The “SHAKE” method (Kr Utler et al. 2015) was used for limiting the bond length of hydrogen atoms, and the “Langevin algorithm” (Larini et al. 2007) was utilized for temperature control, where the collision frequency γ was set to 2 ps−1. The system pressure was 1 atm, the integration step size was 2 fs, and the trajectory was recorded every 10 ps. In addition, “MM-GBSA” (Genheden and Ryde 2015; Hou et al. 2011; Rastelli et al. 2010) was applied to assess optimal binding free energies using the following formulas: where ΔEvdw represents the van der Waals energy and ΔEelec represents the electrostatic energy. The total binding free energy (ΔGbind) is the energy difference between the complex bound state (ΔGcomplex) and the individual free receptor (ΔGreceptor) and the ligand (ΔGligand) as stated in the equation. GGB is an electrostatic contribution to solvation, whereas GSA is a non-polar contribution to solvation. ΔGGB was calculated using the GB model developed by Nguyen et al. (Nguyen et al. 2013) (igb = 2). ΔGSA was detected based on the surface tension (γ) multiplied by the solvent accessibility surface area (SA) [ΔGSA = 0.0072 × ΔSASA (Weiser et al. 1999). Animal experiments Eighty specific pathogen-free (SPF) grade male 6-week-old C57BL/6 mice were obtained from Shanghai Jiesijie Laboratory Animal Co., Ltd. (Certificate No. SCXK [Shanghai] 2023–0004). All experimental protocols were approved by the Animal Ethics Review Committee of Zunyi Medical and Pharmaceutical College (Approval No. ZMPC-2503–005) in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Mice were randomly allocated to experimental groups using a computer-generated randomization sequence (random.org). Group allocation was performed by an independent investigator not involved in behavioral testing or data analysis. All behavioral assessments were conducted by experimenters blinded to group assignment. The sample size (n = 5 per group) was determined based on previous studies employing similar CUMS protocols and behavioral outcome measures, which detected significant antidepressant effects with comparable group sizes. The mice were randomly divided into 8 groups (n = 5 per group): Control group Control group Chronic Unpredictable Mild Stress (CUMS) model group Chronic Unpredictable Mild Stress (CUMS) model group Deacetyl asperulosidic acid group (10 mg·kg⁻1·d⁻.1) Deacetyl asperulosidic acid group (10 mg·kg⁻1·d⁻.1) Deacetyl asperuloside methyl ester group (10 mg·kg⁻1·d⁻.1) Deacetyl asperuloside methyl ester group (10 mg·kg⁻1·d⁻.1) Procyanidin B1 group (20 mg·kg⁻1·d⁻.1) Procyanidin B1 group (20 mg·kg⁻1·d⁻.1) Procyanidin B2 group (20 mg·kg⁻1·d⁻.1) Procyanidin B2 group (20 mg·kg⁻1·d⁻.1) Procyanidin C1 group (20 mg·kg⁻1·d⁻.1) Procyanidin C1 group (20 mg·kg⁻1·d⁻.1) Fluoxetine group (10 mg·kg⁻1·d⁻.1) Fluoxetine group (10 mg·kg⁻1·d⁻.1) Except for the control group, all other groups were subjected to a 35-day CUMS protocol. The stressor schedule was randomly generated using a Latin square design to ensure unpredictability, with two different stressors applied daily in a pseudo-randomized sequence: restraint stress (2 h in a 50 mL conical tube), horizontal shaking (10 min at 100 rpm), cage tilting (12 h at 45° angle), damp bedding (12 h with water-soaked bedding), overnight illumination (12 h continuous light), white noise (4 h at 85 dB), strobe light exposure (2 h at 1 Hz), food and water deprivation (12 h), or cold swimming (5 min in 4 °C water). Stressors were counterbalanced across groups and no identical stressor pair was repeated on consecutive days. From day 22 to day 35 of the CUMS procedure, drug administration was initiated. The doses of each compound were selected based on previous pharmacokinetic and pharmacodynamic studies of structurally related natural products (Manach et al. 2004; Noda et al. 2018; Qu et al. 2014a; Yin et al. 2024a), as well as preliminary dose-finding experiments (data not shown), to ensure therapeutic relevance and translational potential. Fluoxetine was administered via oral gavage to mimic clinical oral administration. This route difference should be considered when comparing the magnitude of effects between PR compounds and the positive control. Fluoxetine group received oral gavage of fluoxetine dissolved in saline. Control and model groups received equal-volume saline (vehicle control). All drug administrations were performed at 9:00 AM daily following the stressor schedule. Behavioral assessments Sucrose preference test (SPT) The 4-day SPT protocol included: Day 1: Two bottles of 1% sucrose solution provided for 24 h adaptation; Day 2: One bottle replaced with distilled water; bottle positions swapped every 12 h to eliminate lateral bias; Day 3: Water deprivation for 24 h; Day 4: Simultaneous access to 1% sucrose solution and distilled water for 24hSucrose preference was calculated as: (sucrose consumption/total fluid intake) × 100%. Open field test (OFT) Locomotor activity was assessed in a 50 × 50 × 40 cm apparatus with a central 25 × 25 cm zone defined as the center area. Mice were placed in the center and recorded for 5-min locomotor activity. Total distance traveled and time spent in center zone were analyzed during the last 4 min using EthoVision XT tracking software. Tail suspension test (TST) Mice were suspended 1 cm from the tail tip in a sound-attenuating chamber. Immobility duration was recorded during the last 4 min of the 6-min test session using a digital timer. Immobility was defined as the absence of all movements except those required for respiration. Forced swimming test (FST) The test was conducted in a 30 cm deep cylindrical tank with water temperature maintained at 22–25 °C. Immobility time was recorded during the last 4 min of the 6-min test period. Immobility was defined as floating without struggling, with only minimal movements to keep the head above water. Sample collection After behavioral testing, mice were anesthetized in a closed chamber saturated with ether vapor and subsequently euthanized via cervical dislocation. Cardiac perfusion was performed with phosphate-buffered saline (PBS) to remove blood. The brain tissues were rapidly harvested on ice, and hippocampal regions were dissected and immediately placed into sterile cryotubes. Samples were flash-frozen in liquid nitrogen and stored at −80 °C until further analysis. ROS measurement in hippocampus by flow cytometry Reactive oxygen species (ROS) levels were quantified using the ROS Assay Kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China). Hippocampal tissues were digested with 0.25% trypsin at 37 °C for 30 min, followed by filtration through a 70 μm cell strainer to obtain single-cell suspensions. The DCFH-DA probe was diluted according to the manufacturer's instructions (negative control without probe, positive control treated with 50 μM H₂O₂). Cells were incubated with the probe at 37 °C for 30 min, washed with PBS, and analyzed using a BD LSRFortessa flow cytometer. The complete gating strategy was as follows: (1) a forward scatter-area (FSC-A) versus side scatter-area (SSC-A) plot was used to identify the main cell population and exclude debris; (2) a forward scatter-height (FSC-H) versus FSC-A plot was applied to exclude doublets and select single cells; (3) fluorescent events were then gated on a histogram of FITC fluorescence intensity (excitation/emission at 488/525 nm), with the negative control sample used to establish the fluorescence threshold; (4) the positive gate was set at the upper 2% of the negative control distribution, and the percentage of DCF-positive cells as well as the median fluorescence intensity (MFI) were recorded for each sample. A minimum of 10,000 single cells were acquired per sample. Fluorescence intensity was quantified to determine intracellular ROS levels. Inflammatory cytokines, oxidative stress markers, and Fe2⁺ analysis in hippocampus Hippocampal tissues were weighed and homogenized in ice-cold 0.7% PBS (1:9 w/v) using a tissue homogenizer on ice. Homogenates were centrifuged at 10,000 × g for 10 min at 4 °C, and supernatants were collected for analysis. Levels of 5-HT, IL-1β, IL-6, TNF-α, GSH, MDA, and SOD activity were measured using commercial kits (Nanjing Jiancheng Bioengineering Institute). Fe2⁺ concentrations were quantified with the Iron Assay Kit (Beyotime, Nantong, China) according to the manufacturer’s protocol. All procedures were performed strictly following kit instructions. Western blot Hippocampal tissues were rapidly dissected on ice and homogenized in ice‑cold RIPA lysis buffer supplemented with a protease and phosphatase inhibitor cocktail (Roche). Homogenates were sonicated for 15 s on ice, incubated at 4 °C for 30 min, and centrifuged at 12,000 × g for 15 min at 4 °C. The supernatant was collected, and protein concentrations were determined using a BCA assay kit (Thermo Fisher Scientific) with bovine serum albumin as the standard. Equal amounts of protein (30 µg per lane) were separated by 10% SDS‑polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride (PVDF) membranes (Millipore) via a wet transfer system. Membranes were blocked with 5% non‑fat dry milk in Tris‑buffered saline containing 0.1% Tween‑20 (TBST) for 1 h at room temperature and then incubated overnight at 4 °C with primary antibodies against ACSL4 (1:1000, ab155282, Abcam), SLC7A11 (1:1000, ab175186, Abcam), GPX4 (1:2000, ab125066, Abcam), EPAS1/HIF‑2α (1:1000, ab199729, Abcam), and GAPDH (1:5000, #5174, Cell Signaling Technology) as the loading control. After thorough washing with TBST, membranes were incubated with horseradish peroxidase‑conjugated secondary antibodies (1:5000, Cell Signaling Technology) for 1 h at room temperature. Immunoreactive bands were visualized using an enhanced chemiluminescence substrate (ECL, Millipore) and captured with a ChemiDoc imaging system (Bio‑Rad). Statistical analysis Data were analyzed using GraphPad Prism 8 (GraphPad Software, USA) and presented as mean ± standard deviation. Homogeneity of variance was confirmed using Levene’s test. Group comparisons were performed by one-way ANOVA followed by Tukey’s post hoc test for multiple comparisons. Student’s t-test was used for two-group comparisons. Statistical significance was defined as P < 0.05. Results Collection of target genes of PR in treating depression The UPLC-Q-TOF–MS analysis was performed to identify the active ingredients in PR. Figure 1A illustrates the ES mode of the total ion chromatograms. The following six active ingredients were identified: deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester l, Procyanidin B1, asperulosidic acid, Procyanidin B2, and Procyanidin C1. The Chromatographic fingerprint of PR water extract as in Fig. 1B. Although the asperuloside content identified in PR was relatively low, we have previously demonstrated that asperuloside could suppress the progression of depression (Yin et al. 2024b, 2025b). Thus, seven active ingredients (the six identified by UPLC-Q-TOF–MS plus asperuloside based on our prior published evidence of its antidepressant activity) were selected for subsequent computational analyses. This apparent discrepancy arises because asperuloside was below the detection threshold in the current UPLC-Q-TOF–MS analysis but was included in the network pharmacology and docking analyses based on established literature support. In addition, the ferroptosis score differed significantly between the depression and control groups (P = 0.014; Fig. 1C). In addition, seven active PR ingredients were included in the computational analyses, and 335 targets of the seven active ingredients were predicted (Fig. 1D). After differential expression analysis, 3,125 DEGs were screened between the depressive and healthy samples (Figs. 1E-F). Fisher’s exact test was applied to explore the correlation between DEGs and potential targets of the active ingredients of PR, and 48 overlapping genes were identified (Table 1). Collection of target genes of PR in treating depression. A UPLC-Q-TOF–MS total ion chromatograms of extracts from PR. B The difference in the ferroptosis score between depression and healthy groups. C Network of active ingredients of PR and targets. Volcano plot D and heatmap E of DEGs between depression and healthy samples Table 1Fisher exact test to explore overlapped genes between DEGs and potential targets of active ingredients of PRDEGsIntersectionNo-intersectionSumP value (fisher.test)Targets of active ingredientsIntersection48287335 < 2.2e-16No-intersection3077030773125287 Fisher exact test to explore overlapped genes between DEGs and potential targets of active ingredients of PR Acquisition of ferroptosis-related genes against depression The value of “power” was selected when the correlation coefficient reached 0.85, which was “power” = 2 (Fig. 2A) to maximize the fulfillment of the scale-free network distribution premise. Next, 25 modules were integrated with minModuleSize = 70 (Fig. 2B). Furthermore, the turquoise module revealed the strongest and most significant association with depression (r = 0.79, P = 2e-05) and ferroptosis score (r = 0.64, P = 0.002), which contained 8,738 genes (Fig. 2C). Overlapping genes between 3,125 DEGs, 8,738 genes in the turquoise module, 565 ferroptosis-related genes, and 335 potential targets of the active PR ingredients were identified, and six ferroptosis-related genes against depression were obtained (Fig. 2D). In addition, the active components of six ferroptosis-related genes against depression were predicted, and 18 active component–target interactions were obtained (Fig. 2E). Enrichment analysis of six ferroptosis-related genes in depression was conducted (Figs. 3A-D), and these genes were involved in 1,117 Gene Ontology (GO) terms and 46 Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. We acknowledge that enrichment analysis using a small gene set (six genes) can yield a large number of terms, potentially reflecting enrichment inflation. Therefore, these results should be interpreted cautiously as exploratory and hypothesis-generating rather than definitive. We have prioritized the most biologically relevant pathways (ferroptosis, HIF-1 signaling, and MAPK cascade) for subsequent experimental validation. Acquisition of ferroptosis-related genes against depression. A Hierarchical clustering of samples and selection of the weight parameter “power” of the adjacency matrix and the mean connectivity. The red line indicates the standard line where the R2 of the correlation coefficient reaches 0.85. B Tree diagram for module division. The module membership is represented by colors in the clustering dendrogram with genes. C Relationships between module and trait. Each cell contains the corresponding correlation and p-value. The table is color-coded for correlation. D Venn diagram of six ferroptosis-related genes against depression. E Prediction of the active components of six ferroptosis-related genes against depression Enrichment analysis. GO-BP A, CC B, MF C terms of six ferroptosis-related genes against depression involved. D KEGG pathways of six ferroptosis-related genes against depression involved. The size of the nodes denotes the number of selected genes in pathways or biological processes, whereas the color of the nodes refers to the significance of the results Identification of key diagnostic genes Eight machine learning model algorithms revealed that “RF” had good predictive performance (AUC value = 0.958), which was selected as the optimal machine learning model (Table 2). Next, the three feature genes, EPAS1, MAPK14, and VEGFA, identified using “RF” were obtained as candidate diagnostic genes. Notably, these three candidate diagnostic genes were both upregulated in depression (all P < 0.01, Fig. 4A) and were significantly and positively correlated with depression and the ferroptosis score (Fig. 4B). Among these three candidate diagnostic genes, the ROC curves revealed that EPAS1 had the highest AUC value (0.925); thus, it was considered a key diagnostic gene (Fig. 4C). Moreover, EPAS1 expression was upregulated in patients with depression in the validation set (P < 0.05; Fig. 4D). Table 2Statistical results of the performance of eight machine learning models in the GSE76826 datasetModelAccuracySensitivitySpecificityAUCLASSO0.8180.6001.0000.800Ridge0.9090.9000.9170.908ElasticNet0.9090.9170.9000.908SVM_RFE0.8640.8330.9000.867RF0.9550.9171.0000.958GBM0.9091.0000.8000.900XGBoost0.5451.0000.0000.500PLSLR0.8180.9170.7000.808 Statistical results of the performance of eight machine learning models in the GSE76826 dataset Identification of key diagnostic genes. A The expression of three candidate diagnostic genes. B Correlation between three candidate diagnostic genes and depression/ferroptosis score. C Receiver operating characteristic (ROC) curves of three candidate diagnostic genes. D Expression of EPAS1 in the validation set. E Four active components of EPAS1. *P < 0.05, **P < 0.01, and ****P < 0.0001 Molecular docking After the active components of the key diagnostic genes were acquired, four active components of EPAS1 were obtained (Fig. 4E), namely, asperuloside, asperulosidic acid, deacetyl asperulosidic acid, and deacetyl asperulosidic acid methyl ester l. In addition, a molecular docking approach was employed where proteins corresponding to the EPAS1 acted as receptors interacting with the four active components: asperuloside, asperulosidic acid, deacetyl asperulosidic acid, and deacetyl asperulosidic acid methyl ester l. These demonstrated affinity toward EPAS1 with respective binding energies of − 8.05 kcal/mol, − 7.471 kcal/mol, − 8.737 kcal/mol, and − 7.360 kcal/mol, respectively (Figs. 5A-D and Table 3). Notably, while deacetyl asperulosidic acid exhibited the most favorable docking score, molecular dynamics simulation-based MM/GBSA analysis revealed that asperuloside and deacetyl asperulosidic acid methyl ester displayed comparable or slightly more favorable binding free energies (− 35.22 and − 35.26 kcal/mol, respectively) compared to deacetyl asperulosidic acid (− 27.78 kcal/mol), suggesting that docking scores alone may not fully capture the dynamic binding behavior. These complementary computational approaches collectively support the multi-target binding potential of PR constituents to EPAS1. The Swiss ADME database was used to evaluate the drug activities of the four active ingredients (Table 4). Molecular docking analysis with key diagnostic genes. A Docking of asperuloside with EPAS1 protein. B Docking of asperulosidic acid with EPAS1 protein. C Docking of deacetylasperulosidic acid binding to EPAS1 protein. D Docking of deacetylasperulosidic acid methyl ester l with EPAS1 protein Table 3Binding energy in molecular dockingTarget_nameLigand_nameDocking_score(kcal/mol)6 × 28-EPAS1Asperuloside−8.056 × 28-EPAS1Asperulosidic_acid−7.4716 × 28-EPAS1Deacetylasperulosidic_acid−8.7376 × 28-EPAS1Deacetylasperulosidic_acid_methyl_ester−7.36 Binding energy in molecular docking Table 4Drug ability of the four active ingredients based on SwissADME databaseActive ingredientsMolecular weightNumber of hydrogen bond acceptorsNumber of hydrogen bond donorsLipophicityMolar refractionAsperuloside414.36114−1.6689.71Asperulosidic acid432.38126−2.4193.47Deacetyl asperulosidic acid390.34117−2.8683.73Deacetyl asperulosidic acid methyl ester404.37116−2.6288.05 Drug ability of the four active ingredients based on SwissADME database Molecular dynamics simulation The RMSD plot revealed the compound and ligand of residual deviation of EPAS1-asperuloside, EPAS1-asperulosidic acid, EPAS1-deacetyl asperulosidic acid, and EPAS1-deacetyl asperulosidic acid methyl ester l (Figs. 6A-B). The RMSD values were small (below 2.5 Å) and the fluctuations were extremely stable, especially for the three docked complexes of the EPAS1-asperuloside, EPAS1-deacetyl asperulosidic acid, and EPAS1-deacetyl asperulosidic acid methyl ester l. In addition, among all proteins in the complex, except for the local protein region, the RMSF of these proteins was low (Fig. 6C), suggesting a highly rigid primary structure of the protein, which could be attributed to binding to small molecules. Molecular dynamics simulation. The root mean square deviation (RMSD) plot revealed the compound A and ligand B of residual deviation of docked complexes. C The root mean square fluctuation (RMSF) plot revealed the fluctuations of docked complexes Binding free energies of interactions Hydrogen bonding is one of the strongest non-covalent interactions. We monitored the number of hydrogen bonds between ligand molecules and proteins during a 100 ns molecular dynamics simulation. Among these four protein–ligand complexes, EPAS1–asperulosidic acid, EPAS1–deacetyl asperulosidic acid, and EPAS1–deacetyl asperulosidic acid methyl ester l formed numerous hydrogen bonds during simulation (Fig. 7A), implying the significant role of hydrogen bonds in the formation of these complexes. Moreover, as shown in Table 5 and Fig. 7B, the ΔGbind of EPAS1–asperuloside, EPAS1–asperulosidic acid, EPAS1-deacetyl asperulosidic acid, and EPAS1-deacetyl asperulosidic acid methyl ester l were − 35.22 ± 4.08, − 25.03 ± 3.72, − 27.78 ± 2.29, and − 35.26 ± 2.86 kcal/mol, respectively. In addition, the MM/GBSA energy decomposition technique was used to identify the top 10 amino acids involved in the binding of small molecules to proteins (Fig. 7C). These amino acids, such as PHE 244 and THR 321, are crucial for the binding of molecules and proteins. Binding free energies of interactions. A The number of hydrogen bonds between ligand molecules and proteins during a 100 ns molecular dynamics simulation. B MM/GBSA binding energy of protein–ligand complexes. C The top 10 amino acids contribute to the binding of small molecules and proteins Table 5Binding free energies and energy components predicted by MM/GBSA (kcal/mol)SystemΔEvdWΔEelecΔGGBΔGSAΔGbindEPAS1/Asperuloside−41.62 ± 2.61−41.24 ± 9.0553.78 ± 7.23−6.13 ± 0.20−35.22 ± 4.08EPAS1/Asperulosidic-acid−38.61 ± 2.7870.88 ± 7.14−51.39 ± 6.70−5.91 ± 0.18−25.03 ± 3.72EPAS1/Deacetylasperulosidic-acid−29.90 ± 3.6324.77 ± 9.83−17.02 ± 9.55−5.63 ± 0.10−27.78 ± 2.29EPAS1/Deacetylasperulosidic-acid-methyl-ester−46.21 ± 2.79−30.04 ± 5.2148.14 ± 6.43−7.15 ± 0.23−35.26 ± 2.86 Binding free energies and energy components predicted by MM/GBSA (kcal/mol) Effects of major components of PR on behavioral parameters in cums-induced depressed mice Compared with the control group, the model group exhibited significantly decreased sucrose preference (Fig. 8A), reduced central exploration times in the open field test (Fig. 8B), and prolonged immobility durations in both the tail suspension (Fig. 8C) and forced swimming tests (Fig. 8D), confirming successful CUMS modeling. Fluoxetine (positive control) significantly reversed these behavioral impairments by increasing sucrose preference (Fig. 8A) and central exploration times (Fig. 8B), while decreasing immobility durations (Fig. 8C-D). Similarly, deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, Procyanidin B2, and Procyanidin C1 significantly improved sucrose preference (Fig. 8A) and reduced swimming immobility time (Fig. 8D). Notably, deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, and Procyanidin B2 increased central exploration times (Fig. 8B), while deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, and Procyanidin C1 decreased tail suspension immobility (Fig. 8C). Among these, Procyanidin B1 demonstrated the most robust antidepressant-like effects, closely resembling the therapeutic profile of fluoxetine. Effects of major components of PR on behavioral parameters in cums-induced depressed mice. A Sucrose preference of mice in eight group. B central exploration times in the open field test of mice in eight group. Immobility durations in the tail suspension C and forced swimming tests D of mice in eight group. *p < 0.05, ***p < 0.001 vs Blank group. #p < 0.05, ##p < 0.01, ###p < 0.001 vs Model group. N = 5. Each experiment was conducted for three times Effects of major components of PR on inflammatory factors in cums-induced depressed mice The model group showed significantly reduced 5-HT levels (Fig. 9A) and elevated pro-inflammatory cytokines, including IL-1β (Fig. 9B), IL-6 (Fig. 9C), and TNF-α (Fig. 9D), compared to controls, confirming CUMS-induced neuroinflammation. Fluoxetine effectively reversed these inflammatory alterations. Furthermore, deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, Procyanidin B2, and Procyanidin C1 significantly reduced IL-1β (Fig. 9B) and TNF-α (Fig. 9D) levels, while deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, and Procyanidin C1 decreased IL-6 concentrations (Fig. 9C). Notably, only Procyanidin B1 significantly restored 5-HT levels (Fig. 9A). These findings highlight Procyanidin B1 as the most effective compound in mitigating neuroinflammatory responses, with effects comparable to fluoxetine. Effects of major components of PR on inflammatory factors in cums-induced depressed mice. The 5-HT levels A, IL-1β B, IL-6 C and TNF-α (Fig D) levels in mice in eight group were detected by corresponding kits. ***p < 0.001 vs Blank group. #p < 0.05, ###p < 0.001 vs Model group. N = 5 Each experiment was conducted for three times Effects of major components of PR on oxidative stress markers in cums-induced depressed mice The model group exhibited significantly reduced hippocampal SOD activity (Fig. 10A) and elevated MDA levels (Fig. 10B), confirming CUMS-induced oxidative damage. Fluoxetine ameliorated these oxidative stress markers. Similarly, deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, Procyanidin B2, and Procyanidin C1 significantly lowered MDA levels (Fig. 10B), while deacetyl asperulosidic acid methyl ester I and Procyanidin B1 enhanced hippocampal SOD activity (Fig. 10A). Effects of major components of PR on oxidative stress markers in cums-induced depressed mice. The SOD activity A and MDA contents B in mice in eight group were detected by corresponding kits. ***p < 0.001 vs Blank group. #p < 0.05, ###p < 0.001 vs Model group. N = 5. Each experiment was conducted for three times Effects of major components of PR on ferroptosis-related markers in cums-induced depressed mice CUMS induced ferroptosis-like damage, evidenced by elevated hippocampal ROS levels (Fig. 11A), decreased GSH concentrations (Fig. 11B), and increased Fe2⁺ accumulation (Fig. 11C) in the model group versus controls. Fluoxetine reversed these ferroptotic alterations. Similarly, deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, Procyanidin B2, and Procyanidin C1 significantly reduced ROS (Fig. 11A) and Fe2⁺ levels (Fig. 11C), while deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, and Procyanidin B2 restored GSH concentrations (Fig. 11B). In addition, the protein levels of GPX4 and SCL7A11 were decreased, while ACSL4 and EPAS1 were increased in the cums-induced depressed mice. Deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester I, Procyanidin B1, and Procyanidin B2 significantly increased the protein levels of GPX4 and SCL7A11, and decreased the protein levels of ACSL4 and EPAS1. While, Procyanidin C1 only decreased the ACSL4 levels, and showed no effects on EPAS1, GPX4 and SCL7A11 protein expressions (Fig. 11D). Effects of major components of PR on ferroptosis-related markers in cums-induced depressed mice. The ROS A, GSH B and Fe2 + C contents in mice in eight group were detected by corresponding kits. D The GPX4, ACSL4, SCL7A11 and EPAS1 protein levels were detected by western blot. ***p < 0.001 vs Blank group. ##p < 0.01, ###p < 0.001 vs Model group. N = 5. Each experiment was conducted for three times Discussion This study aimed to elucidate the antidepressant mechanisms of Psychotria rubra (PR) through an integrative approach combining network pharmacology, molecular docking, molecular dynamics simulation, and in vivo validation in a CUMS-induced mouse model. The key findings are: (1) EPAS1, MAPK14, and VEGF-A were identified as candidate ferroptosis-related diagnostic genes in depression; (2) multiple PR active ingredients, particularly asperuloside, asperulosidic acid, and deacetyl asperulosidic acid, demonstrated favorable binding affinities to EPAS1 in both docking and molecular dynamics analyses; (3) among the tested PR compounds, Procyanidin B1 showed the most potent antidepressant-like effects in vivo, comparable to fluoxetine; and (4) PR active ingredients significantly ameliorated depressive-like behaviors while reducing hippocampal neuroinflammation, oxidative stress, and ferroptosis markers. Depression is a common mental disorder characterized by persistently low mood, reduced interest, and lack of energy. In severe cases, it may affect an individual's cognitive, behavioral, and physiological functions (Mccarron et al. 2021). Our integrative analyses suggest that PR may exert antidepressant effects, potentially involving modulation of ferroptosis-related pathways with EPAS1 as a candidate target. In addition, Procyanidin B1 demonstrated the most robust antidepressant-like activity among the tested compounds, consistent with its reported anti-inflammatory and antioxidant properties. Ferroptosis is a form of programmed cell death driven by iron-dependent lipid peroxidation, and its abnormal activation may contribute to the pathology of depression (Liang et al. 2025). Mechanisms related to ferroptosis, such as oxidative stress imbalance, neuronal damage, and neuroinflammation, could be implicated in the occurrence and development of depression (Feng et al. 2025; Zhu et al. 2024). This evidence, supporting the effects of ferroptosis on the development of depression, is consistent with our findings. In the present study, the ferroptosis score was higher in the depression group than in the healthy group, suggesting that ferroptosis may be a novel target for the prevention and treatment of depression. We used a combination of machine learning and WGCNA and identified three feature genes using “RF” as candidate diagnostic genes, namely EPAS1, MAPK14, and VEGFA. These genes were significantly and positively correlated with both depression and ferroptosis scores. EPAS1, also known as HIF-2α, regulates the transition from acute to chronic hypoxia, subsequently affecting the expression of inducible nitric oxide synthase (Li et al. 2015). Elevated levels of inducible nitric oxide synthase are associated with major depressive disorder (Inserra et al. 2019). In addition, Gao et al. reported that EPAS1 could be a critical biomarker for predicting the overall survival in patients with depression (Gao et al. 2025). MAPK14 is a member of the mitogen-activated protein kinase (MAPK) family that plays a crucial regulatory role in cellular stress responses, inflammatory signal transduction, cell proliferation, differentiation, and apoptosis (Madkour et al. 2021; Wang et al. 2025). MAPK14 coordinates the cellular response to stress and inflammation and plays a significant role in mediating neuronal cell death (Gu et al. 2022). An appropriate level of MAPK14 is crucial for neuronal survival and function and for maintaining the health and plasticity of ischemic neurons (Alam and Scheper 2016). In addition, VEGFA mRNA levels in the whole blood of patients with depression are higher than those in controls, especially in patients with psychotic unipolar depression (Ryan and Mcloughlin 2018), which is in line with the results of this study. Cui et al. have demonstrated the involvement of miRNA-27a in the pathology of depression by mediating the expression of VEGFA (Cui et al. 2018). Thus, these findings indicate that EPAS1, MAPK14, and VEGFA are promising targets for the diagnosis and treatment of depression. The active ingredients of PR were explored to determine their therapeutic effects on depression. Seven active PR ingredients were identified: deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester l, Procyanidin B1, asperulosidic acid, Procyanidin B2, Procyanidin C1, and asperuloside. Asperoside is an iridoterpene glycoside with known antioxidant and anti-inflammatory effects on numerous inflammatory diseases. Yin et al. confirmed that asperuloside repressed the depression progression via the Wnt3α/GSK-3β pathway (Yin et al. 2024b). Morinda officinalisHow can accelerate immune function in depression treatment, and deacetyl asperulosidic acid is the primary component of Morinda officinalisHow (Zhang et al. 2018). In addition, Qu et al. reported that deacetyl asperulosidic acid methyl ester, an antidepressant, can pass through the blood–brain barrier and have targeted distribution with complete pharmacokinetics in the brain (Qu et al. 2014b). Moreover, the ROC curves revealed that EPAS1 had the highest AUC value and therefore it was considered a crucial diagnostic gene. We demonstrated that asperuloside, asperulosidic acid, deacetyl asperulosidic acid, and deacetyl asperulosidic acid methyl ester l interacted with EPAS1. Moreover, the molecular docking analysis highlighted the effectiveness of EPAS1 docking with asperuloside, asperulosidic acid, deacetyl asperulosidic acid, and deacetyl asperulosidic acid methyl ester l, further corroborating these assertions. Furthermore, molecular dynamics simulations validated the predictions obtained from molecular docking and UPLC-Q-TOF–MS identification of these active ingredients within PR, solidifying their therapeutic relevance. Further studies validated that PR may alleviate depression progression by modulating EPAS1 expression. Subsequently, we established a murine model of depression to further evaluate the effects of PR’s major bioactive compounds on depressive-like behaviors, hippocampal inflammation, and ferroptosis. Six compounds were identified: deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester, Procyanidin B1, asperulosidic acid, Procyanidin B2, and Procyanidin C1. Notably, asperulosidic acid was found to be a hydrolysis product of aucubin during extraction and isolation, precluding definitive attribution of biological activity to either compound. To resolve this ambiguity, we selected deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester, Procyanidin B1, Procyanidin B2, and Procyanidin C1 for in vivo validation. Our results demonstrated that most compounds alleviated depressive-like behaviors in mice, as evidenced by increased sucrose preference, enhanced center entries in the open-field test, and reduced immobility time in the tail suspension and forced swim tests. Among these, Procyanidin B1 exhibited the most potent antidepressant effects, closely resembling the therapeutic efficacy of the positive control fluoxetine. Ferroptosis is driven by ROS, whose accumulation initiates lipid peroxidation, leading to membrane damage and cell death. GSH, a critical antioxidant, suppresses lipid peroxidation via glutathione peroxidase 4 (GPX4)-mediated reduction of lipid hydroperoxides, thereby counteracting ferroptosis (Li et al. 2022). MDA, a terminal product of lipid peroxidation, exacerbates membrane disruption and oxidative stress, further promoting ferroptotic progression (Arzuk and Armagan 2024). SOD indirectly modulates ROS homeostasis by converting superoxide (O₂⁻) into hydrogen peroxide (H₂O₂). However, excess H₂O₂, in the presence of Fe2⁺, undergoes Fenton reactions to generate highly reactive hydroxyl radicals, which amplify lipid peroxidation and ferroptosis (Gan et al. 2024; Zhang et al. 2025a, b). These molecules collectively regulate redox balance, and their dysregulation is central to ferroptosis. In vivo studies revealed that the five PR-derived compounds enhanced SOD activity and GSH levels while reducing MDA and ROS concentrations. Combined with their inhibitory effects on Fe2⁺ accumulation, these findings indicate that PR’s active constituents suppress ferroptosis in the hippocampus of depressed mice. Notably, Procyanidin B1 demonstrated superior anti-inflammatory and anti-ferroptotic efficacy, paralleling the effects of fluoxetine. This study further revealed that PR alleviates depression, potentially through modulating EPAS1 expression. Collectively, the five active compounds (deacetyl asperulosidic acid, deacetyl asperulosidic acid methyl ester, Procyanidin B1, Procyanidin B2, and Procyanidin C1) mitigated disease progression in depressive mice, with Procyanidin B1 showing the most robust therapeutic profile, closely mirroring fluoxetine's efficacy. It should be noted that the molecular docking and molecular dynamics simulations presented in this study were specifically designed to predict direct interactions with EPAS1, which was identified as a top diagnostic gene for ferroptosis‑related depression via machine learning and WGCNA. Accordingly, only compounds that showed favorable binding energies toward EPAS1 (i.e., the iridoid glycosides, including deacetyl asperulosidic acid and asperuloside) were highlighted in that section. In contrast, Procyanidin B1, despite exhibiting the most robust antidepressant‑like effects in vivo, did not show particularly strong docking scores with EPAS1. This discrepancy is not unexpected, as Procyanidins are known to exert neuroprotective and anti‑ferroptotic effects through multiple EPAS1‑independent mechanisms, such as activation of the Nrf2/GPX4 axis, direct scavenging of reactive oxygen species, and inhibition of NF‑κB‑mediated neuroinflammation. Therefore, the identification of Procyanidin B1 as the most efficacious compound in the CUMS model is based on its overall in vivo performance, not on its binding affinity to EPAS1. Our study does not claim that Procyanidin B1 acts through EPAS1; rather, it suggests that different active ingredients of PR may contribute to depression relief via distinct or synergistic pathways. Future studies should systematically investigate the direct molecular targets of Procyanidin B1, for example using thermal proteome profiling or chemical proteomics, to fully decipher its mechanism of action. Limitations and future directions Several limitations of this study should be acknowledged. First, the bioinformatics analysis was performed on peripheral blood transcriptome datasets, whereas the in vivo validation focused on hippocampal markers. While blood-based gene expression signatures can reflect systemic changes in depression, the translational relationship between blood-derived targets and hippocampal outcomes requires cautious interpretation. Future studies should validate EPAS1, MAPK14, and VEGFA expression in hippocampal tissue. Second, different routes of administration (intraperitoneal for PR compounds versus oral for fluoxetine) may affect bioavailability. Third, the small training set size (n = 22) for machine learning raises concerns about overfitting despite cross-validation. Future studies should address these limitations through EPAS1 expression validation, canonical ferroptosis markers, dose–response analyses, pharmacokinetic profiling, and investigation of downstream signaling pathways. Conclusion In summary, this study integrates network pharmacology, molecular docking, molecular dynamics simulation, and in vivo experiments in a CUMS-induced mouse model to explore the potential antidepressant effects of Psychotria rubra (PR) and its major bioactive compounds. Our findings show that PR-derived compounds, particularly Procyanidin B1, ameliorate depressive-like behaviors, reduce neuroinflammation, oxidative damage, and ferroptosis-related markers in the hippocampus of depressed mice. Machine learning and correlation analyses suggest that EPAS1, MAPK14, and VEGFA may serve as potential diagnostic or mechanistic targets, with EPAS1 showing the strongest association. Molecular docking and dynamics simulations further indicate that deacetyl asperulosidic acid and other iridoids could bind to EPAS1. However, these conclusions are primarily based on associative and preclinical evidence. The proposed molecular mechanisms, including the direct modulation of EPAS1 by PR components and the causal role of ferroptosis in mediating the antidepressant effects, require further validation using target-specific knockouts, chemical inhibitors, or clinical samples. Moreover, the translational relevance of the observed effects to human depression remains to be established. Therefore, while this study provides a strong rationale and testable hypotheses for future research, the current evidence should be interpreted as exploratory rather than definitive. Future studies should focus on direct target engagement, causal pathway analysis, and evaluation in more advanced preclinical models or clinical settings. Acknowledgements Not applicable. Authors' contributions All authors participated in the design, interpretation of the studies and analysis of the data and review of the manuscript. L Y drafted the work and revised it critically for important intellectual content; C L, J Y, Y Z and S M were responsible for the acquisition, analysis and interpretation of data for the work; H W made substantial contributions to the conception or design of the work. All authors read and approved the final manuscript. Funding The study was supported by: Zunyi Science and Technology Plan Program, HZ(2024)400, Li Yin, Zunyi Medical College PhD Start-up Fund, zyyz-bsqd(2025)03, Li Yin, Yulin City Scientific Research and Technology Development Program, 202432217, Chengshu Lu, Research Team on the Conservation and Utilization of Genuine Medicinal Materials in Northern Guizhou, HZ(2023)454, Ju Ye, National Natural Science Foundation of China, 81960658, Li Yin, Zunyi City Science and Technology Innovation Talent Team Construction Project, Zunshi KCTD (2025)No.90, Zunyi Technology Innovation Center, Zunyi Kehe CXZX(2025)No.6, Research Team of Zunyi Medical College, (2026) No. 03. Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Animal study All experimental protocols were approved by the Animal Ethics Review Committee of Zunyi Medical and Pharmaceutical College (Approval No. ZMPC-2503–005). All animal experiments should comply with the ARRIVE guidelines. All methods were carried out in accordance with relevant guidelines and regulations. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.