The active components of EE were extracted from the TCMSP (https://tcmspw.com/tcmsp.php). The compounds were screened according to oral bioavailability (OB) ≥30% and drug likeness (DL) ≥0.18 (16). SDF format of the 2D structures of the compounds was obtained by entering the CAS numbers of EE in PubChem (https://pubchem.ncbi.nlm.nih.gov). Then it was imported into STP (http://swisstargetprediction.ch/) to obtain the predicted targets of the chemical components. The active ingredient targets were obtained after screening by probability>0. Target information from the STP and TCMSP database was integrated to get the potential targets of EE components (17). UniProt (https://www.uniprot.org) was used to convert each target into genes (18). EE effects on osteoclast for potential target procurement. OC-related targets were obtained from the GeneCards (https://www.genecards.org), OMIM (https://www.omim.org), and TTD databases (http://db.idrblab.net/ttd/) with osteoclast differentiation and osteoclastogenesis as the keywords. GeneCards were filtered above the median relevance score (19). The intersection of EE components and OC-related targets was plotted using an online Venn analysis tool.

Protein-protein interaction (PPI) underlies most biological processes in living cells and is essential for understanding cellular physiology in normal and disease states (20). The potential target genes of EE action on osteoclast were uploaded to the STRING database (https://cn.string-db.org/). The protein species was set to “Homo Sapiens” and the interaction score was set to 0.4. Meanwhile the unconnected nodes in the network were hidden to obtain and the interaction results were saved as TSV format files. The files were imported into Cytoscape software to draw PPI network diagrams, and topological parameters were analyzed by applying the network analyzer of Cytoscape’s built-in network analysis plug-in (21). According to the degree of target genes, the greater the degree was, the more critical role it played in the process of EE acting on osteoclast.

To further elaborate the mechanism of EE action on osteoclast, Metascape (https://metascape.org) was used to perform GO term enrichment analysis and KEGG pathway enrichment analysis on the intersection targets (22). Microsign online graphing software was used to visualize GO functional enrichment and KEGG pathway analysis data. The species option was set to “Homo Sapiens”, and the analysis about the cell components, molecular functions, biological process, and signaling pathways involving the targets was conducted. Pathways and biological processes for which P < 0.05 were extracted.

The obtained core targets were molecular docking to the core components. The protein structures of the core targets were downloaded from the PDB database (https://www.rcsb.org/), and then the core proteins were dehydrated and delighted with PyMol 2.4.0 software. The structures of the core components were retrieved by PubChem, downloaded in SDF format, and converted to pdb format with OpenBabel 3.1.1. AutoDock 4.2.6 software was used to routinely process protein receptors and small molecule ligands and save them in pdbqt format. Molecular docking and binding energy calculation scripts were run using AutoDock Vina 1.1.2 (23), and some results were visualized using PyMol software (24).

EE was purchased from CHENGDU MUST BIO-TECHNOLOGY CO.,LTD. Twenty C57BL/6J female mice (7–week-old; 18 ± 1g) were acquired from the Guangdong Medical Laboratory Animal Center. These animals were placed in a 12 hours light/dark cycle. After acclimation for 7 days, all mice were randomly divided into four groups (n=5 per group): Sham group, OVX group, EE-L (20 mg/kg) group, and EE-H (40 mg/kg) group. All animals had free access to food and water. EE was dissolved in 1% carboxymethyl cellulose sodium (CMC-Na). Conversely, the Sham and OVX group received intragastric administration of equal volume CMC-Na. After 12 weeks, mice were sacrificed and collected their left femur, serum and feces. The Ethics Committee of Guangdong Pharmaceutical University had reviewed and approved the experimental protocol according to the Guide for the Care and Use of Laboratory Animals.

Micro-CT (SCANCO uCT-100 detector, Switzerland) was used to scan the morphological structure of the left distal femur of mice, and the bone trabeculae in the area 0.8 mm above 0.5 mm from the growth plate were selected as the region of interest (ROI) for 3D reconstruction. Bone mineral density (BMD), bone volume fraction (BV/TV), bone surface/bone volume (BS/BV), number of trabeculae (Tb.N) and trabecular separation (Tb.Sp) of the ROI were analyzed with evaluation analysis software. The analysis ranges were kept consistent for all specimens.

The serum was obtained from anesthetized animals by retroorbital puncture at the end of the study. The levels of TRAP, CTX, P1NP, TNF-α, LPS and IL-6 were measured by ELISA according to the manufacturer’s instructions.

After collecting the mice feces, the feces were placed in sterile cryopreservation tubes, rapidly frozen in liquid nitrogen, and stored in an ultra-low temperature refrigerator at -80°C for gut microbiota diversity analysis. The E.Z.N.A.®soil DNA kit (Omega Bio-Tek, Norcross, GA, U.S.) was used to extract DNA from each mouse fecal sample, and the extracted DNA was detected by 1% agarose gel electrophoresis. Then all primers were designed based on the conserved region. 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) were used for PCR amplification of the V3-V4 region of the 16S rRNA gene to perform the illumina deep sequencing. PCR products were detected by 2% agarose gel electrophoresis. The purified PCR products were sequenced using the NEXTFLEX®Rapid DNA-Seq kit. The PCR products were sequenced through Illumina’s Miseq PE300 platform (Shanghai Majorbio Biopharm Technology, China).

All data were represented as mean ± SD. The ANOVA test was used for the data with normal distribution, and the Mann–Whitney test was used to analyze data that did not meet the assumptions of the ANOVA (SPSS, version 26.0). When P<0.05, the differences were considered to be statistically significant.

The 2D structure of EE is shown in Figure 1A . We retrieved 15 herbal component targets through the TCMSP and STP databases. All the targets were transformed into 883 genes using the UniProt database. 1152 target genes related to osteoclast differentiation and osteoclastogenesis were captured from the GeneCards, OMIM, and TTD databases. 130 potential targets of EE action on osteoclast were selected by taking the overlap through the Venn online platform ( Figure 1B ).

The interaction relationships of potential EE targets affecting osteoclast were obtained from the STRING database ( Figure 1C ), in which the PPI network displayed 130 functional proteins with 1686 interactions. With the node degree as the evaluation parameter, where a higher node degree indicates a more important role in the PPI network. The top 10 core targets in terms of the node degree are TP53, AKT1, JUN, CTNNB1, STAT3, HIF1A, EP300, CREB1, IL1B, and ESR1 ( Figure 1D ).

The biological process, cellular composition, and molecular function were selected by GO analysis of 130 common targets. The GO enrichment analysis showed that those selective targets are mainly involved in processes of transcription factor binding (GO:0008134), response to hormone (GO:0009725), gland development (GO:0048732), response to growth factor (GO:0070848) ( Figure 2A ). To further explore the potential pathways of EE in the treatment of OP, KEGG analysis showed that common targets are mainly enriched in the Wnt signaling pathway (has:04310), HIF-1 signaling pathway (has:04066), cAMP signaling pathway (has:04024) and Calcium signaling pathway (has:04020) ( Figure 2B ).

To further look into the molecular basis of this compound and identify associated mechanistic targets, EE was selected to perform molecular docking with the top 10 core targets predicted by the PPI network analysis. The absolute values of the docking score indicate the affinity of the components with the targets and the stability of the component-target conformation. An absolute value greater than 5.0 indicates a good binding, and greater than 7.0 indicates a strong binding ( Table 1 ) (25, 26). The binding energy of target proteins and the active compounds is approximately between −5.0 and −7.0 kcal/mol, which means that EE bind well to 10 core target proteins. Comprehensive analysis shows that the docking scores of EE with HIF1A have the highest absolute value among the targets. We chose the top 4 target proteins macromolecules and small compound molecules with the best docking affinity for visualization by PyMol, which were HIF1A, EP300, STAT3 and CREB1 ( Figure 3 ).

The reconstructed 3D images confirmed that OVX underwent significant bone loss, which was attenuated by EE treatment ( Figure 4A ). The values of BMD, BV/TV, BS/BV, and Tb.N were decreased significantly in the OVX compared with the Sham, although the value of Tb.Sp was increased significantly (P<0.05, P<0.01). The levels of BV/TV, BS/BV, and Tb.N were increased significantly, while Tb.Sp was presented an opposite trend in EE-H (P<0.05). The value of BMD was enhanced but not significantly different (P>0.05) ( Figures 4B–F ).

Serum P1NP levels of mice in the OVX was significantly decreased compared with the Sham (P<0.01), a sharp increased in P1NP by EE (P<0.05), moreover, the contents of TRAP, CTX, TNF-α, LPS and IL-6 were significantly increased in the OVX (P<0.05, P<0.01). Treatment of EE reversed the increased serum levels of TRAP, CTX, TNF-α, LPS and IL-6 (P<0.05, P<0.01) ( Figures 4G–L ).

We analyzed the number of common and unique OTUs among the different groups to show that four groups had different compositions ( Figure 5A ). The Shannon index on OTU level indicated that the amount of sequencing data was large enough to reflect the vast majority of gut microbial diversity information in samples ( Figure 5B ). The principal coordinates analysis (PCoA) and non-metric multidimensional scaling (NMDS) with bray_curtis distances on OTU level revealed a distinct clustering of the microbiota composition for each group ( Figures 5C, D ). Furthermore, ACE and Chao index were used to characterize community species richness and Simpson index were used to characterize community diversity. Compared with the OVX, EE could reduce the ACE and Chao index, while improving the Simpson index value. This indicated that EE treatment increased the community diversity of the gut microbiota and diminished community richness ( Figures 5E–G ).

To study the specific changes in bacterial communities, cluster histograms were drawn to show the changes in gut microbiota at the class, family, genus and species levels in each sample ( Figures 6A–D ). The relative abundances of the dominant bacterial families, genus and species were also compared among the groups ( Figures 6E–G ). At the class level, Bacteroidia, Bacilli, and Clostridia were the predominant in all groups. At the family level, compared to the Sham, OVX reduced the relative abundance of Lactobacillaceae, while the relative abundance of Clostridiaceae significantly increased. Compared to the OVX, EE increased the relative abundance of Lactobacillaceae and decreased the relative abundance of Clostridiaceae. At the genus level, OVX decreased the relative abundance of Lactobacillus significantly, while increased the relative abundance of Clostridium_sensu_stricto_1. EE treatment increased Lactobacillus and decreased Clostridium_sensu_stricto_1 significantly. At the species level, the contents of Uncultured_bacterium_g_norank_f_Muribaculaceae and Uncultured_bacterium_g_Clostridium_sensu_stricto_1 in the OVX were higher than those in the Sham, whereas the contents of Lactobacillus_murinus decreased. Compared to the OVX, EE reduced the relative abundance of Uncultured_bacterium_g_norank_f_Muribaculaceae and Uncultured_bacterium_g_Clostridium_sensu_stricto_1, while the relative abundance of Lactobacillus_murinus significantly increased.