The following databases were used in the network pharmacology analysis: Bioinformatics Analysis Tool for Molecular Mechanism of Traditional Chinese Medicine (BATMAN-TCM; http://bionet.ncpsb.org.cn/batman-tcm/); Traditional Chinese Medicine Systems Pharmacology (TCMSP; https://tcmsp-e.com/), SwissTargetPrediciton (http://swisstargetprediction.ch/); Pharmmapper (http://lilab-ecust.cn/pharmmapper/index.html); GeneCards (https://www.genecards.org/), PubChem (https://pubchem.ncbi.nlm.nih.gov/); Online Mendelian Inheritance in Man (OMIM; https://www.omim.org/); DisGeNET (https://www.disgenet.org/home/); GeneCards (https://www.genecards.org/); Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/), UniProt (https://www.uniprot.org/); Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/); and Search Tool for the Retrieval of Interacting Genes/Proteins (STRING; https://string-db.org/). The following software programs were used: Cytoscape v3.8.2 (https://cytoscape.org/); Rx64 v4.1.0 (https://cran.r-project.org/); and Biocondcutor (https://www.bioconductor.org/).

The Simplified Molecular Input Line Entry System (SMILES) and International Chemical Identifier key (InChiKey) for Danshensu were retrieved from the PubChem database. Danshensu was searched in BATMAN-TCP (prediction score of the target gene ≥20; adjusted p value of the prediction result <0.05), TCMSP, SwissTargetPrediction, and Pharmamapper databases based on the results of the search.

The keyword “ankylosing spondylitis” was used as input in the OMIM, GeneCards, and DisGeNET databases to identify genes related to AS. The original microarray data of GSE25101 and GSE73754 were analyzed with GEO2R to identify differentially expressed genes (DEGs) and generate a volcano plot; according to log₂ fold change (FC), the −log₁₀ (adjusted p values) were divided into upregulated genome, downregulated genome, no statistical difference, and non-expressed difference genome (screening conditions: adjusted p value <0.01, |log₂ FC|≥1.0). Both AS-related gene and DEG names were normalized and merged and duplicate items were deleted, and the obtained gene sets were used as disease targets.

All targets were searched in the UniProt database with the species set to human, and the collected protein or gene information was corrected to the standard UniProt name (Symbol) of the target. The set of Danshensu targets was mapped to the set of disease targets and intersected to obtain the potential targets of Danshensu in AS.

Topologic properties, degree, and interaction of the distribution network from the STRING database were analyzed using the cytoHubba plugin of Cytoscape.

The drug component–disease target gene names were encoded by UniProt IDs. GO functional annotation of the targets and KEGG pathway enrichment analysis were performed using Rx64 v4.1.0 software with bioc-Lite.R, clusterProfliter.R, and pathview.R packages from the Bioconductor platform.

Cytoscape v3.8.2 software was used to construct the disease–drug component–target–signaling pathway interaction network, in which the nodes represented the disease, drug component, target, and signaling pathways and edges represented interactions between drug component and target, target and AS, or target and signaling pathway. The topologic properties of the network were analyzed using the Network Analysis plugin of Cytoscape.

Hip capsular ligaments were collected from three AS patients at Guangdong Provincial People’s Hospital. The study was approved by the ethics committee of the hospital. Fibroblasts were extracted from the hip capsular ligaments as previously described (Qin et al., 2018). Briefly, ligaments were washed twice with phosphate-buffered saline (PBS) and then cut into small pieces that were placed in 10-mm culture dishes. After 3–4 h, Dulbecco’s Modified Eagle’s Medium (DMEM; Gibco, Shanghai, China) containing 15% fetal bovine serum (FBS; Gibco) and 1% penicillin–streptomycin (Solarbio, Beijing, China) was added to the culture dishes. After cells had adhered, the ligament pieces were removed and the medium was replaced with fresh medium containing 10% FBS and 1% penicillin–streptomycin. The medium was changed every 3–4 days and cells from passage 3–5 were used for experiments.

Primary osteoblasts from the skull of 24-h newborn Sprague–Dawley rats were provided by Guangdong Medical University. Briefly, the calvaria was cut into small pieces and digested at 37°C with trypsin for 15 min, followed by 0.1% collagenase I for 20 min. The supernatant was centrifuged at 800 rpm for 10 min, and the cell precipitate was collected and inoculated in 25-cm² culture vials while the supernatant was transferred back to the centrifuge tube containing the bone fragments for additional digestion. The above steps were repeated 3 times. The cells were grown in DMEM containing 10% FBS and 1% penicillin–streptomycin at 37°C in a humidified atmosphere of 5% CO₂. The culture medium was refreshed every 2–3 days. When the cells reach 80% confluence, they were digested with 0.25% EDTA–trypsin (Solarbio) and passaged in a 1:2 or 1:3 ratio. Cells from passage 3–5 were used for experiments.

The 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay was used to evaluate the effect of Danshensu on osteoblast and AS fibroblast proliferation. The cells were seeded in 96-well plates at a density of 3×10³ (osteoblasts) and 5×10³/well (fibroblasts), respectively. After 24 h, the cells were divided into nine groups (5 wells per group) that were treated with Danshensu at a concentration of 0, 0.1, 0.3, 1, 3, 10, 30, 100, or 300 μM. After incubation for 7 days, 20 μl MTT (5 mg/ml) was added to each well. The mixed medium was discarded 4 h later and replaced with 150 μl dimethylsulfoxide (Solarbio), and the plate was agitated for 20 min at room temperature. A microplate reader (Bio-Rad Laboratories, Hercules, CA, United States) was used to measure the optical density (OD) of each well at a wavelength of 570 nm.

Cells were washed twice with PBS and fixed with 4% paraformaldehyde for 30 min. After three washes with PBS, the cells were stained with ALP detection solution for 30 min at room temperature according to instructions of the ALP kit (Leagene, Beijing, China). To measure ALP activity, cells were lysed overnight at 4°C with radioimmunoprecipitation assay (RIPA) lysis buffer (Beyotime, Shanghai, China), followed by incubation for 15 min at 37°C in ALP detection buffer from the ALP activity kit (Jiancheng, Nanjing, China). The reaction was terminated and coloring solution was added to each well. The OD was measured at 520 nm with a microplate reader. The protein concentration of each well was measured with a bicinchoninic acid protein assay kit (Beyotime). ALP activity is reported as U/100 g of protein.

Cells were washed twice with PBS and then fixed with 70% ethanol for 20 min at room temperature; they were then washed with water and stained with 0.2% alizarin red (Sigma, Tokyo, Japan) solution (pH 4.1) for 30 min. The solution was discarded and the cells were washed 3 times with distilled water.

After culturing for 14 days, total RNA was extracted from cells grown in 6-wells plates using TRIzol reagent (Ambion, Austin, TX, United States). The RNA was reverse transcribed using the PrimeScript RT reagent kit (Takara, Dalian, China) and qPCR was carried out to detect mRNA levels of Runt-related transcription factor (RUNX)2, osterix (SP7), osteocalcin (OCN), bone morphogenetic protein (BMP)2, and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (internal control). Cycling conditions were as follows: 95°C for 5 min, followed by 40 cycles of 95°C for 15 s and 60°C for 60 s. Forward and reverse primers were synthesized by Sangon Biotech (Shanghai, China) and are shown in Table 1.

After culturing for 14 days, cells were lysed with RIPA lysis buffer supplemented with phenylmethylsulfonyl fluoride (Beyotime), and 30 μg of protein per well were separated by sodium dodecyl sulfate polyacrylamide gel electrophoresis on a 10% acrylamide gel and electrotransferred to a polyvinylidene difluoride membrane that was incubated with antibodies against GAPDH (1:2000; Abway, Shanghai, China); RUNX2 (1:1000), collagen type I (COL1; 1:1000), and SP7 (1:1000) (all from Abcam); OCN (1:500; Affinity, Cincinnatin, United States); and JNK (1:1000) and extracellular signal-regulated kinase (ERK; 1:1000) (both from Signalway Antibody, College Park, MD, United States). Enhanced chemiluminescence substrate (Beyotime) was used to detect the protein signal.

All data are presented as mean ± SD. SPSS v22.0 software was used to analyze the data. One-way analysis of variance was used to assess differences between groups, with a least significant difference post hoc test when groups had equal variance (p > 0.05) and Dunnett’s post hoc test in the alternative case. Differences with a p value < 0.05 were considered statistically significant.

We identified 21 putative targets of Danshensu (Figure 2A) in BATMAN-TCM along with 22 in TCMSP, 100 in SwissTargetPrediction, and 273 in Pharmmapper (Figure 2B). The identified genes were aggregated and duplicates and inconsistently named genes were removed, leaving 406 genes as potential targets.

Using the keyword “ankylosing spondylitis,” 108 AS-related genes were identified in OMIM along with 2,274 in GeneCards and 710 in DisGeNET. A total of 324 and 61 DEGs were obtained from GSE25101 and GSE73754, respectively (Figure 2D). The identified genes were aggregated and duplicates and inconsistently named genes were removed, leaving 2,944 AS-related genes (Figure 2C).

AS-related genes and Danshensu targets were intersected in a Venn diagram, which yielded 47 drug component–disease targets (Figure 3A). The 47 genes were imported into the STRING database; the functions of multiple proteins were selected and a combined score >0.4 was used as the threshold while disregarding isolated nodes to obtain a PPI network for Danshensu AS targets. The topologic analysis of the network revealed 164 connections among 47 targets (Figure 3B).

Cytoscape was used to visualize key genes as a network from which overlapping genes were selected for further analysis by determining the node degree or interaction of sequenced candidate genes using the cytoHubba application. The top 10 hub genes in the network of Danshensu AS targets were IL6, ESR1, PTGS2, EGFR, IL10, TGFB1, TIMP1, MMP1, NR3C1, and RELA (Table 2 and Figure 3C).

The target gene names (symbol) were transformed into gene IDs by searching the UniProt database, and the biocLite.R package of Bioconductor was used to perform GO functional annotation and KEGG pathway enrichment analysis of the target genes using Rx 64 v4.0.1 software, with p = 0.05 and Q = 0.05 as the threshold values. The top 40 GO terms are shown in Figure 4. Most were related to inflammatory response, oxidative stress, the MAPK cascade, etc. The KEGG pathway enrichment analysis yielded 13 signaling pathways among which ERK and JNK signaling pathways were highly enriched, suggesting that they are key pathways for the anti-osteogenic effect of Danshensu (Figure 4).

The drug component–disease target, active ingredient molecular ID, and drug component–disease–signaling pathway interaction data were imported into Cytoscape, and a drug component–target interaction network was generated using the Merge function in the software (Figures 3D,E). These targets were presumed to be key targets of Danshensu in AS.

To investigate the effect of Danshensu on osteoblast and AS fibroblast proliferation, the cells were exposed to increasing concentrations of Danshensu ranging from 0.1 to 300 μM. After culturing for 7 days, the MTT assay was performed. Danshensu did not affect the proliferation of osteoblasts and AS fibroblasts at concentrations between 0.1 and 30 μM, while at 100 μM osteoblast proliferation was significantly inhibited and at 300 μM, both osteoblast and AS fibroblasts were affected (Figure 5). We therefore used 1, 3, and 10 μM Danshensu for the following experiment.

We investigated the effect of TNF-α on osteoblast differentiation by ALP staining and activity assay after 7 days of treatment. Both ALP staining and activity were increased in osteoblasts by treatment with 0.01 ng/ml TNF-α for 7 days (Figure 6A). We also examined the effect of TNF-α on mineralization in osteoblasts by ARS. There were more calcium nodules in the group treated with 0.01 ng/ml TNF-α than in the other groups (Figure 6B). TNF-α at concentrations 1 and 10 ng/ml was not toxic to osteoblasts, and there were no calcium nodules in the 10 ng/ml group.

To confirm the effect of low TNF-α concentrations on osteoblast differentiation, we evaluated the expression of BMP2 and SP7—2 genes related to this process—by qPCR. Application of 0.01 ng/ml TNF-α resulted in the upregulation of BMP2 and SP7 (Figure 6C). We also examined the levels of RUNX2 and COL1A1 proteins by western blotting and found that they were significantly increased in the presence of 0.01 ng/ml TNF-α, whereas a concentration of 10 ng/ml reduced the protein level of RUNX2 (Figure 6D). Based on these results, 0.01 ng/ml TNF-α was used in subsequent experiments to create a proinflammatory environment for inducing osteoblast differentiation.

To investigate the anti-osteogenic effect of Danshensu, osteoblasts were treated with 0.01 ng/ml TNF-α and different concentrations of Danshensu, and ALP staining and activity were evaluated 7 days later. Danshensu inhibited ALP activity in TNF-α–induced osteoblasts in a dose-dependent manner. ALP staining was also reduced in cells treated with Danshensu compared to those treated with 0.01 ng/ml TNF-α only (Figure 7A). Similar results were obtained by ARS (Figure 7B). Additionally, qPCR analysis showed that RUNX2, BMP2, and SP7 mRNA expression was reduced in cells treated with Danshensu compared to those treated with TNF-α only (Figure 7C) and after 14 days, cells exposed to 0.01 ng/ml TNF-α and 1, 3, or 10 μM Danshensu had lower levels of COL1A1 and RUNX2 proteins compared to those treated with TNF-α only (Figure 7D).

To determine whether Danshensu affects normal osteoblasts, we assessed the osteogenic potential of osteoblasts treated with 1, 3, and 10 μM of Danshensu. The results of ALP staining and the activity assay showed that Danshensu increased osteogenic differentiation (Figure 8A); this was confirmed by ARS (Figure 8B). The qPCR analysis showed that RUNX2, BMP2, and SP7 mRNA expression was increased by Danshensu treatment, an effect that was concentration-dependent (Figure 8C); and a similar effect was observed on COL1A1 and RUNX2 protein levels (Figure 8D).

As fibroblasts are known to contribute to heterotopic ossification, we extracted primary fibroblasts from hip ligaments of AS patients and treated these cells with Danshensu (1, 3, and 10 μM) to further assess whether Danshensu inhibits ectopic ossification in AS. The ALP staining and activity assay showed that Danshensu suppressed osteogenic differentiation of AS fibroblasts (Figure 9A). This was accompanied by decreased expression of COL1 and RUNX2 proteins (Figure 9B) and downregulation of OCN, SP7, and RUNX2 transcripts (Figure 9C) as determined by western blotting and qPCR, respectively. ARS confirmed the results of ALP staining (Figure 9D).

Based on the results of the network pharmacology analysis, we examined the phosphorylation levels of JNK and ERK in AS fibroblasts treated with Danshensu for 14 days. The results showed that p-JNK and p-ERK levels were significantly downregulated in cells treated with 3 and 10 μM Danshensu (Figure 10).