Commercial JGC (specification: 0.5 g per pill) was purchased from Guizhou SSLF Pharmaceutical Co., Ltd. (Guizhou, China, approval number: Z20043621). According to the literature (Sridhar et al., 2021), JGC was powdered and extracted using a Soxhlet extractor with 6 times the amount of 90% ethanol. The solvent was then concentrated using an electrically heated blast drying oven at 45°C. Subsequently, the concentrate was lyophilized with a freeze dryer and weighed. The JGC extract was dissolved in DMSO (20 mg/ml) and stored at −80°C for later use.

The human synovial cell line SW982 was kindly provided by Procell Life Science and Technology Co., Ltd. (Wuhan, China). SW982 cells have been shown to possess characteristic features similar to synovial fibroblasts which makes them an ideal tool to study synovitis in OA (Karuppagounder et al., 2022). The cells were cultured in DMEM/Ham’s F12 medium (DMEM/F12; HyClone, Logan, UT, United States) with 10% fetal bovine serum (PAN Biotech, Aidenbach, Germany) and 1% penicillin/streptomycin (Gibco, Grand Island, NY, United States).

The cell proliferative capacity was determined by Cell Counting Kit-8 assays (CCK-8, Biosharp, Guangzhou, China). Cells (10,000/well) were plated in 96-well plates, and DMSO, CTGF or JGC was added according to the experimental design. CTGF is a pro-inflammatory cytokine, that is, upregulated in OA and contributes to synovial hyperplasia (MacDonald et al., 2021). The working concentration of CTGF was 25 ng/ml, and that of JGC was 20 μg/ml. After 24 h, the supernatant was replaced with CCK-8 working solution, and the absorbance at 450 nm was measured.

An Annexin V-FITC Assay Kit (Merck, NJ, United States) was used to detect apoptosis in synovial cells. The cells were plated in 6-well plates (50,000/well) and processed as described above. After 24 h, the cells were dissociated and stained according to the instructions provided with the kit. In brief, cells were digested with trypsin, washed gently with PBS, resuspended in buffer solution to 1 × 10⁶ cells/ml. Then 5 μl Annexin V-FITC was added, and the mixture was incubated in the dark for 5 min 5 μl propidium iodide (PI) was added to the cells before analyzed. We measured the proportion of FITC(+) cells by flow cytometry.

Single-cell sequencing data for synovial cells were downloaded from the GEO database (no. GSE176308), and 10X genomics data were loaded into the R package Seurat (v4.0.2). Synovial cells were obtained from 4 patients with early-stage OA (both painful and non-painful sites) and 4 patients with end-stage OA (painful sites) (Nanus et al., 2021). Cell quality control was applied to remove low-quality cells with less than 300 detected genes or with more than 10% mitochondrial genes. After normalizing the data, the cells were dimensionally reduced and clustered according to the top 2,000 highly variable genes. The FindIntegrationAnchors algorithm found a set of anchors between Seurat objects from different patients. These anchors could be used to integrate the objects using the IntegrateData function. Harmony package (v1.0) was used to remove the batch effect, the diversity clustering penalty parameter was set to 2 and the ridge regression penalty parameter was set to 1.

The dynamic states of synovial cells were assessed using the Monocle algorithm (v2.18.0). Monocle uses an unsupervised algorithm to order whole-transcriptome profiles of single cells and produce a ‘trajectory’ of an individual cell’s progress through differentiation. We applied the “reduceDimension” function to compute the CellDataSet object as a lower dimensional trajectory. The Discriminative Dimensionality Reduction with Trees (DDRTree) method was chosen for its ability to reduce dimensionality while discriminating between different data points. Following dimension reduction, the two features with the most significant amount of information were extracted and used as the coordinate axes to visualize the trajectory. Branched expression analysis modeling (BEAM) was performed to identify genes with branch-dependent expression and thus elucidate fate decision mechanisms.

Independent cell cycle analysis was performed for each synovial cell. The “CellCycleScoring” function in the Seurat package was used to assign cell cycle scores according to S- and G2/M-phase genes, which were identified following procedures described in a previous study (Kan et al., 2022). The number of control features selected from the same bin per analyzed feature was set to 100 and the random seed was set to 1. The cells were classified into G1, S, and G2/M phases based on the maximal score of each cell cycle phase program.

We obtained information regarding the main raw materials from the JGC drug manual. Information about the main active ingredients of these raw materials was obtained from the relevant literature (Supplementary Table S2). The SDF format files of molecular structures were downloaded from the Pubchem database (https://pubchem.ncbi.nlm.nih.gov/). Targets of these molecular structures were predicted using the SwissTargetPrediction database (http://www.swisstargetprediction.ch/) (Daina et al., 2019). The species was confined to “Homo sapiens”, and the predicted targets with a probability more than 0.3 were included in this study.

Macromolecular structures were downloaded from the RCSB PDB database (https://www.rcsb.org), and biological ligands were accessed from PubChem database. PDB files were converted to the PDBQT format. We used AutoDockTools software to search for possible active pockets, removed all water molecules and assigned hydrogen polarities. AutoDock Vina was employed to conduct molecular docking between the active ingredients and targets, then took the conformation with the highest docking score (Affinity). Finally, we used the PyMOL software to visualize the results of molecular docking.

Bilateral tests were performed for all statistical tests. A p-value lower than 0.05 was considered to indicate statistical significance. R software version 4.0.2 (https://www.r-project.org/) was used for the analysis. The following R language packages were used in this study: “dplyr”, “Seurat”, “monocle”, “monocle”, and “iTALK”. The “drug-material-target” network was visualized using Cytoscape_3.7.2 (https://cytoscape.org).

JGC is widely used for OA treatment with ideal clinical efficacy. According to the instructions, the main raw materials of JGC include Cibotium barometz (CB [Cyatheaceae; Cibotium barometz (L.) J. Sm]), Epimedium (ED [Berberidaceae; Epimedium sagittatum (Siebold & Zucc.) Maxim]), Clematidis radix (CR [Ranunculaceae; Clematis chinensis Osbeck]), Zaocys dhumnades (ZD [Colubridae]), Achyranthes bidentata Blume (ABB [Amaranthaceae; Achyranthes bidentata Blume]), Chaenomeles sinensis (CS [Rosaceae; Pseudocydonia sinensis (Dum.Cours.) C.K. Schneid]), Pueraria lobata (PL [Fabaceae; Pueraria montana var. lobata (Willd.) Maesen & S.M. Almeida ex Sanjappa & Predeep]), Curcuma longa (CL [Zingiberaceae; Curcuma longa L., Sp. Pl.: 2 (1753)]), Psoralea corylifolia Linn. (PCL [Fabaceae; Cullen corylifolium (L.) Medik]), and Campanumoea javanica bl (CJB [Campanulaceae; Codonopsis javanica (Blume) Hook. f. & Thomson, Ill. Himal. Pl. t.16 B (1855)]). Certain materials (ED, ABB, CS, PL, CL, and CR) reportedly have significant anti-inflammatory and antioxidant activities, and the aqueous extract of CR exerts a good anti-osteoarthritis effect (Cheng et al., 2013; Lin et al., 2019; Cheng et al., 2020; Jeon et al., 2020; Lin et al., 2021; Razavi et al., 2021). The reasonable compatibility of these materials guarantees curative efficacy.

Synovial tissue shows discordant hyperplasia and inflammation during OA progression. The human synovial cell line SW982 was treated with JGC to assess the effect of this drug on synovial hyperplasia. In normal synovial cells, the inhibition of proliferation by JGC was not significant, indicating tolerable drug toxicity. We then induced hyperproliferation using the growth factor CTGF, and JGC exerted a more pronounced inhibitory effect on the proliferation of active synovial cells (Figure 1A). Flow cytometry showed that the proportion of FITC(+) synovial cells was significantly increased, showing the apoptosis-promoting effect of JGC on SW982 cells (Figure 1B). The inflammatory cytokine IL-1β was applied to induce intense inflammation in synovial cells. Although the expression levels of numerous inflammatory genes (IL-1β, IL-6, IL-8, NOS2, and TNF-α) were clearly increased, JGC treatment significantly reversed the increase in expression caused by inflammatory stimulation (Figure 1C). We also found similar trends for the intracellular reactive oxygen species (ROS) levels: inflammation led to increased ROS levels in SW982 cells, and this increase was relieved after JGC addition (Figure 1D). These results confirm the therapeutic effect of JGC on synovitis in vitro.

To deeply dissect the molecular mechanism of JGC in the treatment of OA synovitis, scRNA-seq data from 4145 synovial fibroblasts (SFs) were examined in this study. SFs were clustered into nine color-labeled subsets based on their unbiased transcriptome signatures (Figure 2A). The cell cluster properties were preliminarily assessed based on cluster-specific markers (Figures 2B,C; Supplementary Figure S1; Supplementary Table S1): the cells in SF-0 expressed high levels of IGFBP6, MFAP5, and SEMA3C, indicating their high proliferative capacity; the cells in SF-1 overexpressed CXCL12 and ID1, suggesting a stronger inflammatory stimulus; the cells in SF-2 expressed MMP2 and WISP2, which play decisive roles in fibrosis; the cells in SF-5 showed relatively high expression of Piezo2, a mechanosensitive channel; the cells in SF-6 expressed RNASE1, indicating decreased adhesion to cartilage; the cells in SF-7 expressed genes critical for synovial angiogenesis (expressing SCUBE3); and the cells in SF-8 expressed relatively high levels of a cell cycle-related gene (CENPM).

We further calculated module scores to assess their inflammatory and proliferative activities, which are the two most prominent pathological features of synovitis. Consistent with the abovementioned results, the SF-1 synovial cells showed the highest level of inflammation, whereas the SF-0 cells exhibited an excessive proliferative capacity (Figures 3A,B). Overall, the proportions of cells from patients with or without pain, according to clinical information, did not significantly differ among the clusters; however, higher proportions of cells in SF-0, SF-1, and SF-2 were obtained from end-stage OA patients (Figures 3C,D). A cell‒cell communication analysis revealed complex ligand‒receptor interactions in the synovial microenvironment, and intercellular crosstalk was mainly divided into cytokines, growth factors and others (Figure 3E). Based on the cytokine categories, the synovial cells in SF-1 expressed higher levels of CXCL12, which interacts with the ITGB1 receptor of surrounding cells to regulate proinflammatory cytokine production (Kong et al., 2020). The growth factor category revealed that CTGF secreted by SF-7 cells interacts with LRP1, which is highly expressed on the surface of cells in other clusters, to induce pathological progression (Schnieder et al., 2020).

The Monocle pseudotime algorithm was used to profile the fate trajectory of synovial cells. The cells were dimensionally descended and arranged in a typical dendritic shape (Figure 4A), and the fate trajectory was divided into three cell states based on bifurcation points (Figure 4B, state 1 to state 3). By comparing the gene patterns in distinct cell states, we found certain classical progenitor/stem cell markers to be significantly overexpressed in cell state 1 (OCT-4, TRA-1-81, SSEA4, NANOG, etc.). Thus, cell state 1 was defined as the origin of the trajectory (Figure 4C), and the synovial cells gradually differentiated into two distinctive fates as the trajectory progressed (Figure 4D).

We screened for “branch-dependent” genes that changed as the cell fate developed and divided these genes into two gene modules. A Gene Ontology (GO) enrichment analysis of “branch-dependent” genes helped annotate the cellular properties across different cell fates (Figure 4E). Certain functions that are beneficial to synovitis recovery were significantly activated in cell fate 1 (e.g., negative regulation of the inflammatory response and cell growth). However, some terms that suggest pathogenesis were enhanced in cell fate 2 (such as positive regulation of angiogenesis). The expression patterns of some canonical synovitis regulators were further assessed, and certain restorative genes (such as NMB, APOE and SMAD7) were highly expressed in cell fate 1 but decreased in cell fate 2. In addition, some pathogenic genes, such as ASPN and ACTA2, showed completely contrary trends (Figure 4F). A cell cycle analysis showed that the proportion of actively proliferating cells (G2/M) was significantly higher in cell fate 2, indicating likely tissue hyperplasia (Figure 4G). What’s more, the two pathways associated with pain (prostanoid and eicosanoid signaling) showed increased activation in cell fate 2, suggesting that these cells were more likely to induce clinical symptoms (Figure 4H). In summary, these results suggest that cells in cell fate 1 contribute to recovery and that cells in cell fate 2 lead to synovitis progression.

A differential expression analysis between the two cell fates identified a total of 403 key synovitis genes, including 195 and 208 upregulated genes in cell fate 1 and cell fate 2, respectively (Figure 5A). Furthermore, by summarizing previous research results, we collected 122 bioactive molecules from the raw materials of JGC (Supplementary Table S2). Subsequently, 151 potential targeting relationship pairs were predicted from the SwissTargetPrediction database (Supplementary Table S3), and a “drug-material-target” network was generated to visualize the potential therapeutic mechanism (Figure 5B). By taking the intersection of JGC targets with key genes of synovitis, five promising functional targets (AKR1C3, VEGFA, CYP1B1, MMP2, and PTGS2) were obtained (Figure 5C). Molecular docking was performed to simulate the interaction between bioactive compounds and binding pockets, which revealed a molecular basis for this picomolar affinity (Supplementary Figure S2). The quercetin-AKR1C3 pair exhibited the best affinity, indicating that this pair constitutes the most promising molecular mechanism (Figures 5D,E; Table 1).

Further PCR results confirmed the hypothesis that AKR1C3 expression was elevated in inflamed synovial cells and effectively inhibited by the addition of JGC (Figure 6A). Rescue experiments were performed to characterize the regulatory relationship. AKR1C3 overexpression significantly attenuated the JGC-induced inhibitory effect on synovial cell proliferation (Figure 6B). Similarly, the anti-inflammatory effect of JGC on synovial cells was clearly counteracted by AKR1C3 overexpression (Figure 6C). Taken together, our findings suggest that JGC treats synovitis in osteoarthritis by inhibiting AKR1C3.