Detailed data on patients with osteoarthritis admitted to the Department of Rheumatology and Immunology of the First Affiliated Hospital of the Anhui University of Chinese Medicine was collected and analyzed, over the period from September 2009 to May 2022. The herbal prescriptions and disease-associated laboratory indices, such as the inflammatory markers: C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR), the immune indices: immunoglobulin A (IgA), immunoglobulin M (IgM), immunoglobulin G (IgG), complement component 3 (C3) and complement component 4 (C4), and the oxidation index: superoxide dismutase (SOD) recorded in electronic medical record systems were routinely abstracted for participants. The research protocol and procedures have been approved by the Ethics Committee of the First Affiliated Hospital of Anhui University of Chinese Medicine (Review No. 2022MCZQ01). This study is a retrospective data mining study based on hospital information systems. Because the system uses secondary level identification to identify information, patient informed consent was not required. In total, 12,842 patients with OA were examined, including 12,226 who received herbal remedies.

Patients were divided into an experimental group (including Coicis Semen in herbal prescriptions) and a control group (not including Coicis Semen in herbal prescriptions). The control group had 3,992 cases, separately, and the experimental group had 8,234 cases.

Association Rule. The compatibility rules of the Coicis Semen/indicators were extracted by SPSS Clementine v.11.1 (IBM Corp., Armonk, NY, USA). Nominations for the application of herbs or indicators were assigned 1, while nominations for non-use are assigned 0. An association rule between X and Y has three degrees of support, confidence, and lift, respectively (Ding et al., 2020). The specific calculation formula of association rules was as documented in previous studies (Fang et al., 2022b; Zhou et al., 2021).

Propensity score matches. The propensity score for each individual was calculated based on factors such as age, sex, body mass index (BMI), length of hospital stays (LOS), and the underlying disease to reduce the bias in treatment selection. The patients in the two groups were precisely matched using the method of propensity score as elucidated by Hanaya Raad and Victoria Cornelius (Raad et al., 2020).

Random walking. The lab-indexed random walk model was evaluated with the ORACLE 10 g tool. The random walk model had long-range correlations, which clearly implies that immuno-inflammatory indicators were valid. The principle and formula of random walking were first proposed by Karl Pearson (Pearson, 1905). Subsequently, it has also been flexibly used in the evaluation of medical clinical data (Fang et al., 2022a; Li et al., 2022; Zhou et al., 2021). The entailed detailed calculation formula is based on previous research (Zhou et al., 2021).

The Traditional Chinese Medicine Systems Pharmacology Database served as the source for the collection of all the composite parts of Coicis Semen. To draw additional conclusions, only compounds that met the criteria suggested by the TCMSP database, such as oral bioavailability (OB) ≥ 30 and drug-likelihood (DL) ≥ 0.18, were chosen as candidate compounds (Chen, 2011). Further, among the compounds with OB < 30 or DL < 0.18, which were explored with “compound (name)” and “arthritis” [all fields] in PubMed databases to discover relevant research, the compounds in purified form centered on anti-OA mechanisms were also regarded as bioactive compounds and included for further study. The chemical structures were drawn by ChemDraw (version 14.0). The underlying target genes of Coicis Semen were then aggregated via the TCMSP platform and selected for the human species. The Coicis Semen target was obtained from the UniProt database.

Information on OA-associated target genes was screened and surveyed by the following electronic databases: GeneCards, PHARMGKB, OMIM, TTD, DrugBank, DisGeNET, and MalaCards. We used the UniProt database to perform target-human gene name conversion for disease targets and delete duplicate targets, which were searched using the keyword “osteoarthritis.” The results of the seven databases analysis were summarized, integrated, and deduplicated to obtain OA-related targets. Venn plots were drawn using the Venny online tool to identify the active ingredient targets and the disease targets for OA. Subsequently, as a result of the intersection of Coicis Semen and OA, the latent target set of Coicis Semen for treating OA was identified.

To set up the “Coicis Semen active component-disease target network” diagram, the screening results were inputted into Cytoscape software (version 3.7.2) (Piñero et al., 2021). Cytoscape can analyze and construct network graphs, and it can set essential parameters including volume, color, shape, etc. The connections between nodes in the network reflect the interactions between them. The higher the degree value, the more nodes are connected to it (Huang et al., 2009).

Protein-protein interaction (PPI) relates to the physical binding of two or more proteins in response to different disturbances and circumstances (Watanabe et al., 2021). A protein interaction network diagram was created with a species that qualified as “Homo sapiens”, at a confidence level of 0.400, and hidden disconnected nodes in the network. Our next step was to evaluate the topological properties of the nodes in the interaction network using the CytoNCA plug-in (Tang et al., 2015): “degree centrality (DC),” “closeness centrality (CC),” “betweenness centrality (BC),” “network centrality (NC),” “eigenvector centrality (EC),” and “local average connectivity (LAC).” We gained the core objectives of the network by measuring the value and nature of nodes, which are measured by the six above parameters (Wood et al., 2008). The higher a node is, the greater its importance.

We used R (Rx64, 3.6.3), and the biological molecular functional annotation system DAVID (https://david.ncifcrf.gov/), to have access to the Coicis Semen treatment of osteoarthritis target set enrichment analysis. Statistical significance was determined by Fisher’s test using p values  < 0.05 and q values < 0.05.

The PDB database was searched by limiting the organism to “Homo sapiens only” to retrieve 3D structures of osteoarthritis target genes. The 2D structure of the Coicis Semen active component was retrieved, and the structure was optimized using Chem3D. AutoDock Tools, a visualization software, was applied to molecular docking. Eventually, the binding model and 3D visual analysis of the docking results were performed using the PyMOL software. LigPlot⁺ software demonstrated the 2D structure of protein-ligand molecular interactions (Schöning-Stierand et al., 2020).

A random sample of 30 patients with OA from the Department of Rheumatology and Immunology at the Anhui Provincial Hospital of Traditional Chinese Medicine was assigned to the experimental group (including Coicis Semen in herbal prescriptions) or the control group (not including Coicis Semen in herbal prescriptions). A total of 15 healthy donors were included in the healthy controls (HC) from the physical examination center. Details of the clinical samples are presented in Suppl. Table S1. The study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Anhui University of Chinese Medicine (Review No. 2022MCZQ01). Informed consent was signed by all participants. Fasting blood was collected in the morning, cell culture supernatants were collected, centrifuged (1000 g, 4°C, 10 min) to remove cell debris, and then stored at −80°C until further quantification. Commercially available ELISA kits for PPAR-α (Genmei Technology Co., Ltd., Wuhan, China), ESR1 (Genmei Technology Co., Ltd., Wuhan, China) and MAPK1 (Genmei Technology Co., Ltd., Wuhan, China) were used to detect the expression according to the instructions in the kit.

Coicis Semen (100 g, S27205, MedChemExpress, Shanghai, China) was dissolved in 100 mM dimethyl sulfoxide (DMSO, Gibco, California, USA) to make a stock concentration. For the experiments, the required concentration of Coicis Semen was obtained by an additional dilution of the concentrated stock solution with culture media. The human chondrocytes C28/I2 (Haodi Huatuo Biotechnology Co., Ltd., Shenzhen, China) were cultured in complete Roswell Parker Memorial Institute (RPMI)-1640 medium (Gibco, California, USA) containing 10% fetal bovine serum (FBS, Gibco, California, USA) at 37°C with 5% CO₂. Lipopolysaccharide (LPS, 10 µg/mL, Sigma-Aldrich, St. Louis, Missouri, USA) was utilized to stimulate C28/I2 cells to establish the model of OA in vitro (Jia and Wei, 2021). LPS has been documented as an initial factor in the regulation of inflammatory response during the pathogenesis of OA (Scotece et al., 2018).

The treatment in the study was divided into the normal control group (NC, normal cultured C28/I2 cells), model group (MC, treated with 10 μg/mL LPS for 12 h), Coicis Semen low-dose group (Coicis Semen-L, treated with 10 μg/mL LPS for 12 h, then treated with 25 μmol/L Coicis Semen for 24 h), Coicis Semen medium-dose group (Coicis Semen-M, treated with 10 μg/mL LPS for 12 h, treated with 50 μmol/L Coicis Semen for 24 h), and Coicis Semen high-dose group (Coicis Semen-H, treated with 10 μg/mL LPS for 12 h, treated with 100 μmol/L Coicis Semen for 24 h). The dosage of Coicis Semen was referred to as the in vitro experimental drug dosage on the MedChemExpress (MCE, Shanghai, China) official website and related literature studies (Du et al., 2021; Zhang et al., 2019a).

Cell treatment with radioimmunoprecipitation assay (RIPA) buffer (Biosharp, Beijing, China) containing protease and phosphatase inhibitors (Biosharp, Beijing, China) was performed to extract total proteins from C28/I2 cells. The primary antibodies used in this study included anti-β-actin (1:1000 dilution, Zs-BIO, Beijing, China), anti-ESR1 (1:1000 dilution, Affinity, Wuhan, China), anti-MAPK1 (1: 2000 dilution, Abcam, Cambridge, England), and anti-PPAR-α (1:2000 dilution, Affinity, Wuhan, China). The secondary antibodies included goat anti-rabbit IgG (1:5000 dilution, Zs-BIO, Beijing, China), anti-β-actin (1:10000 dilution, Zs-BIO, Beijing, China), anti-ESR1 (1:5000 dilution, Affinity, Wuhan, China), anti-MAPK1 (1:5000 dilution, Abcam, Cambridge, England), and anti-PPAR-α (1:5000 dilution, Affinity, Wuhan, China). The relevant instructions were referred to and finally the ECL hypersensitive luminescence kit (Beyotime, Shanghai, China) was used to detect each protein.

The statistical analyses were carried out with SPSS (version 22.0; IBM, NY, USA). Continuous variables were presented as the mean ± standard deviation, or medians and interquartile ranges (IQR). The nonparametric statistical analyses between the two groups were performed using the Wilcoxon signed-rank and Mann–Whitney U-test for paired and unpaired data, respectively. The Student’s t-test was used to compare data that was typically distributed. Categorical variables were expressed as numbers (%) and were compared using the χ² test. Variations were recognized statistically as significant at p < 0.05. To analyze the variances and construct the images, GraphPad Prism 9.0 software (GraphPad, CA, USA) was used.

Based on the analysis of the apriori module, each item is ranked with the highest level of confidence, with a minimum support of 60% and a minimum confidence level of 80%. The relationship between oxidative-inflammatory index optimization and Coicis Semen was analyzed using the association rules. We conclude that Coicis Semen with improved CRP has a confidence level of 89.80%. The improvement in confidence on the ESR was 85.73%, and on the SOD, it was 83.82%. At the same time, all lifts are greater than 1. The OA oxidative-inflammatory index showed substantial improvement with Coicis Semen treatment, as shown in Table 1.

The baseline features for each group are compared before and after propensity score matching in Table 2. We used propensity score analysis to adjust for potential differences in the socio-demographics of patients in the two groups, including age, gender, and BMI. In the analysis of the two groups, comorbidities of interest (coronary heart disease, cerebral infarction, hypertension, chronic gastritis, hyperuricemia, osteoporosis, fatty liver) were significantly different, but that was comparable to that of the matched control group and the matched experimental group. The oxidative-inflammatory indices (ESR, IgA, SOD, CRP, C3, IgM, C4, IgG) were markedly greater in the experimental group than the control group (p < 0.05) but were comparable after propensity-score matching. Subsequently, it could be concluded that ESR, C3, IgA, CRP, IgG, and C4 were considerably reduced after treatment compared to pre-treatment. Compared to the control group, the experimental group had a significant reduction in ESR, C3, IgA, IgG, C4, CRP, and SOD after treatment (Table 3).

The random walking model showed that there was a positive correlation between Coicis Semen treatment and ESR levels, CRP levels, IgA levels, IgG levels, C3 levels, SOD levels, and C4 levels. It is not difficult to find from Table 4 that the experimental group also performs better than the control group in the walking evaluation of different laboratory indicators (ESR, CRP, IGA, IGG, SOD, C3, C4). Fig. 1 visually displays the walking trend of the oxidative-inflammatory index for both groups.

By January 2022, a total of 38 compounds were recognized in Coicis Semen. All of the identified compounds were filtered through Absorption, Distribution, Metabolism, and Excretion (ADME) screening, and a total of 24 active compounds met the criteria, accounting for 63.16% of all 38 compounds. The properties of the compounds are demonstrated in Table 5.

By searching GeneCards, OMIM, TTD, PharmGkb, Disgenet, Drugbank, and Malacards disease databases, we collected 1835, 6, 32, 9, 1827, 150, and 63 target genes related to OA, respectively (Fig. 2A, Suppl. Table S2). On an average, 2924 OA-related targets were obtained after removing duplicated genes. Overall, a Venn graph intersection of identified targets for Coicis Semen and OA (Fig. 2B) identified 61 target genes.

A “herb-ingredient-target” network diagram intuitively revealed the link between the ingredients of Coicis Semen and target genes related to OA-related genes (Fig. 3). The network comprised 130 nodes (108 active compounds and 21 targets) and 224 edges (degree > 1). As shown in Fig. 3, the red triangles represent Coicis Semen, the green rectangle represents the active chemical composition of Coicis Semen, the blue circles represent the target genes of OA, and the lines between them represent the relationship between chemical composition and OA target genes.

An interaction network of 58 nodes and 190 edges (Fig. 4A) was constructed using CytoNCA based on DC, BC, CC, EC, NC, and LAC parameters. The thresholds for the first screening were LAC > 1.714, degree > 5, closeness > 0.137, eigenvector > 0.0626, betweenness > 15.638, and network > 3.05, and the results showed 22 nodes and 111 edges (Fig. 4B). Twenty-two targets were then screened further away. Next, the second screening threshold was degree > 10, eigenvector > 0.214, LAC > 5.8, betweenness > 8.746, closeness > 0.656, and network > 6.654. We obtained an EP300, JUN, ACTB, RELA, MYC, MAPK1, ESR1, and PPARA core PPI network with 8 nodes and 26 edges (Fig. 4C).

By inputting 61 key genes into the DAVID website, 1478 biological processes (BP), 73 cellular components (CC), and 76 molecular functions (MF) were acquired with a 0.05 threshold. These targets were primarily enriched in oxidative stress, inflammatory response, and multi-multicellular organism process in BP, as shown in Fig. 5A, including positive regulation of the inflammatory response (GO: 0050727), positive regulation of oxidative stress (GO: 0006979), regulation of multi-multicellular organism process (GO: 0044706), etc. In CC analysis, the cellular components of these key targets mainly were cytoplasmic vesicle lumen (GO: 0060205), vesicle lumen (GO: 0031983), protein-lipid complex (GO: 0032994), to name a few. The MF analysis predicted that these key targets are compactly connected with transcription coregulator binding activity (GO: 0001221), nuclear receptor activity (GO: 0004879), etc.

The KEGG results revealed that the targets were abundant in 136 pathways, including Hypoxia-inducible factor (HIF-1), IL-17, cyclic adenosine monophosphate (cAMP), TNF, phosphoinositide 3-kinase (PI3K-Akt), MAPK, the NF-kappa B signaling pathway, etc. (Fig. 5B). In the enrichment analysis, it was found that Coicis Semen is capable of acting on OA through multiple pathways, primarily involved in the oxidative-inflammatory response. The Sankey bubble diagram revealed the genes contained in the first 30 pathways (Fig. 5C). The top 3 target genes with the highest frequency were MAPK1, ESR1, and PPARA (Suppl. Table S3).

A molecular docking study was performed in this study to verify the possible mutual effects between eight core target proteins and Coicis Semen active compounds. A heatmap was used to visualize docking scores, as shown in Fig. 6. Lower docking energies indicate higher stable receptor-ligand conformations (Tang et al., 2015). Target proteins were strongly bound to Coicis Semen compounds with binding energies between −4.3 and −10.8 kcal/mol, suggesting that Coicis Semen-derived compounds have a favorable affinity for eight core target proteins. Ultimately, the top four target protein macromolecules and small compound molecules with the best docking affinity were selected for visualization by Pymol (<−10 kcal/mol) (Fig. 7). Interestingly, MAPK1, ESR1, and PPARA were selected as the target proteins, which were also consistent with the enrichment results (Fig. 5C).

The expression of ESR1 and PPAR-α in OA patients was remarkably lower than that in healthy controls, but MAPK1 expression was significantly higher. Compared with pre-therapy, the expression level of ESR1 and PPAR-α increased, while that of MAPK1 decreased in both the control group and experimental group after treatment. When the difference value between the two groups before and after treatment, the expression of ESR1 and PPAR-α in the experimental group increased remarkably more than that in the control group, while MAPK1 expression decreased significantly more (Fig. 8).

Human chondrocytes C28/I2 were cultured and treated with LPS to build an in vitro model of OA. The expression of ESR1 and PPAR-α protein in the MC group was significantly lower than that in the NC group (p-ESR1 = 0.0022, p-PPAR-α = 0.0018). The expression level of ESR1 protein in the Coicis Semen group was significantly higher than that in the MC group (p-Coicis Semen-M = 0.0025, p-Coicis Semen-H = 0.0031) while the expression level of PPAR-α protein in the Coicis Semen group was significantly higher than that in the MC group (p-Coicis Semen-L = 0.0412, p-Coicis Semen-M = 0.0038, p-Coicis Semen-H = 0.0024). The protein expression levels of PPAR-α and ESR1 in the Coicis Semen-H group were higher than those in the Coicis Semen-M group (p-ESR1 = 0.0315, p-PPAR-α = 0.0221). Further, the expression of MAPK1 protein in the MC group was significantly higher than that in the NC group (p-MAPK1 = 0.0015). The expression level of MAPK1 protein in the Coicis Semen group was significantly lower than that in the MC group (p-Coicis Semen-L = 0.0032, p-Coicis Semen-M = 0.0028, p-Coicis Semen-H = 0.0034). The results showed that Coicis Semen significantly decreased the expression of MAPK1 and increased the expression of ESR1 and PPAR-α (Fig. 9).