Skip to main content
NCBI home page
Orthopaedic Surgery logoLink to Orthopaedic Surgery
. 2022 Dec 13;15(5):1333–1347. doi: 10.1111/os.13617

Mining Potential Drug Targets for Osteoporosis Based on CeRNA Network

Zheng Wang 1,2, Xiao‐fei Zhang 3, Mao‐peng Wang 1,2, Shuo Yan 1,2, Zheng‐xu Dai 1,2, Qing‐hang Qian 2, Jie Zhao 1, Xin‐long Ma 3,4,5,✉, Bing Li 1,✉, Jun Liu 1,2,✉
PMCID: PMC10157711  PMID: 36513616

Abstract

Objective

To identify key pathological hub genes, micro RNAs (miRNAs), and circular RNAs (circRNAs) of osteoporosis (OP) and construct their ceRNA network in an effort to explore the potential biomarkers and drug targets for OP therapy.

Methods

GSE7158, GSE201543, and GSE161361 microarray datasets were downloaded from Gene Expression Omnibus (GEO) database. Differentially expressed genes (DEGs) were identified by comparing OP patients with healthy controls and hub genes were screened by machine learning algorithms. Target miRNAs and circRNAs were predicted by FunRich and circbank, then ceRNA network were constructed by Cytoscape. Pathways affecting OP were identified by functional enrichment analysis. The hub genes were verified by receiver operating characteristic (ROC) curve and real time quantitative PCR (RT‐qPCR). Potential drug molecules related to OP were predicted by DSigDB database and molecular docking was analyzed by autodock vina software.

Results

A total of 179 DEGs were identified. By combining three machine learning algorithms, BAG2, MME, SLC14A1, and TRIM44 were identified as hub genes. Three OP‐associated target miRNAs and 362 target circRNAs were predicted to establish ceRNA network. The ROC curves showed that these four hub genes had good diagnostic performance and their differential expression was statistically significant in OP animal model. Benzo[a]pyrene was predicted which could successfully bind to protein receptors related to the hub genes and it was served as the potential drug molecules.

Conclusion

An mRNA‐miRNA‐circRNA network is reported, which provides new ideas for exploring the pathogenesis of OP. Benzo[a]pyrene, as potential drug molecules for OP, may provide guidance for the clinical treatment.

Keywords: Circular RNA, Competing Endogenous Network, Drugs, Machine Learning, MicroRNA, Osteoporosis

In order to explore the pathogenesis of osteoporosis and provide a new perspective for the treatment of osteoporosis, we constructed a ceRNA network and predicted a drug molecule of osteoporosis.

graphic file with name OS-15-1333-g009.jpg

Introduction

Osteoporosis (OP) is a common systemic bone disease, which is characterized by bone loss, decreased bone mineral density, deterioration of bone microstructure, and increased bone brittleness, resulting in an increased risk of fracture. 1 The main pathological changes of OP are increased activity of osteoclasts or decreased activity of osteoblasts. 2 The pathogenesis of OP is complex, and the currently known signaling pathways affecting osteoporosis include Wnt/β‐catenin signaling pathway, Sema3A/NRP1/PlexinA1 signaling pathway, and RANKL/OPG signaling pathway, etc. 3 , 4 , 5 With the aging of the population, the number of osteoporosis patients is gradually increasing. According to statistics, in 2019, across Europe (the European Union, plus Switzerland and the UK), an estimated 32 million people over the age of 50 suffered from OP, which was equivalent to 5.6% of the total population over the age of 50 in Europe. 6 Fractures caused by OP seriously impacted the life quality of people, increased the mortality rate of elderly patients, and also aggravated the social burden even further. 7 , 8 Currently, the main treatment methods for OP are drug therapy, such as estrogen, bisphosphonate, etc., but their side effects and poor oral absorption cannot be ignored. 9 Another study showed that monoclonal antibody teriparatide may be effective for the treatment of osteoporotic fractures, but it was still controversial. 10 Therefore, clarifying the pathogenesis of OP at the molecular level and developing related drugs are becoming urgent requirements.

More and more studies had shown that non‐coding RNA, miRNA, and circRNA played important roles in the occurrence and development of OP. MicroRNA can affect the occurrence of OP by altering the differentiation of osteoblasts or osteoclasts. According to the research, MiR‐155 could affect the function of osteoclasts by affecting AMPK signal pathway and Wnt signal pathway, and inhibiting the expression of miR‐155 could improve OP. 11 MiR‐96 could promote osteoblast differentiation and bone formation in strong mice by activating Wnt signal pathway. 12 Meanwhile, circRNA could inhibit OP by affecting the role of miRNA. For example, circIGSF11 could decrease osteoblast differentiation by competitively binding miR‐199b‐5p. 13 CircSIPA1L1 up‐regulated the expression of ALPL and promoted the differentiation of osteoblasts through binding with miR‐204‐5p. 14 In the mechanism of miRNA binding to circRNA or mRNA, miRNA acts as a post‐transcriptional regulator by binding to the microRNA response elements (MREs) of the target mRNA. 15 However, circRNA suppresses its effect through MREs competitive combination with miRNA, which is the essence of ceRNA network. 16 Therefore, exploring ceRNA network plays an important role in the diagnosis and treatment of OP.

During this study, we downloaded the OP and normal tissue microarray data from the GEO database. The objectives included the following three points. Firstly, we analyzed the open microarray dataset in order to screen the differential expressed genes of OP by comparing with healthy samples. Scientific machine learning algorithms were used to screen hub genes which are of great importance for the diagnosis of OP. And then these hub genes were used to predict miRNA and circRNA related to OP to construct the ceRNA network. Moreover, the biological pathways and functions of hub genes and target miRNAs were analyzed. Furthermore, we used these hub genes to predict and evaluate small molecular drugs related to OP therapy. Our results achieved the above research purposes and provided guidance for the clinical treatment of OP in the future.

Methods and Materials

Data Acquisition

Osteoporosis and control datasets, GSE7158, GSE201543, and GSE161361, were downloaded from Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo). GSE7158 included 14 controls and 12 patients with OP, six cases of OP and four controls included in GSE201543, while three cases of OP and three controls included in GSE161361.

Data Pre‐processing

The data dimension reduction function of the “factoextra” package (https://cran.r-project.org/web/packages/factoextra/index.html) in R was used to perform the principal component analysis (PCA), and the cluster analysis was also performed on GSE7158. The “hgu133plus2.db” packet (http://www.bioconductor.org/packages/hgu133plus2.db/) was used for ID conversion to convert the probe into a genetic symbol for the next step of research. Duplicate genes and genes with missing values were removed.

Screening of DEGs

Differential expressions of genes between the OP and control datasets were screened using the “LIMMA” package in R (http://www.bioconductor.org/packages/release/bioc/html/limma.html). DEGs are presented in the form of a volcano figure and heat map. A value of p < 0.05 and |log2 FC| > 1 were considered to show statistical significance for differential expression.

Screening and Verification of Hub Genes

The three algorithms were adopted for screening of hub genes for OP, including random forests (RF), least absolute shrinkage, and selection operator (LASSO) logistic regression and support vector machine‐recursive feature elimination (SVM‐RFE). The random forest algorithm was performed with “randomForest” package (https://www.stat.berkeley.edu/~breiman/RandomForests/) in R. This study carried out LASSO logistic regression investigation with R package “glmnet” (https://cran.rproject.org/web/packages/glmnet/index.html). The SVM‐RFE classifier was constructed using the R package “e1071” (http://www.cran.r-project.org/web/packages/e1071) with five‐fold cross‐validation. The online tool VENNY (https://bioinfogp.cnb.csic.es/tools/venny/) was used to select the overlapping genes, which were defined as hub genes. Receiver operating characteristic (ROC) curves were established by “pROC” package (http://www.cran.r-project.org/web/packages/pROC/index.html) of R to evaluate the predictive value of hub genes. The diagnostic effect in distinguishing control samples from OP samples was evaluated by area under the curve (AUC) values.

Identification of the Target miRNAs and Target cirRNAs

GEO2R tool was used to analyze GSE201543 and GSE161361 to obtain the differentially expressed miRNAs (DEMs) and differentially expressed circRNAs (DECs). The FunRich (http://www.funrich.org/) and TargetScan (https://www.targetscan.org/vert_80/) databases were used to predict miRNAs with hub genes regulatory activity. Cytoscape (Version 3.9.0), a network biology analysis and visualization tool, was used to visualize the result. By using the online tool VENNY, overlapping miRNAs between the predicted miRNAs of hub genes and the DEMs were defined as target miRNAs. The circRNAs which can be regulated by target miRNAs were predicted by circbank (http://www.circbank.cn/searchMiRNA.html). And overlapping circRNAs between the predicted circRNAs of hub genes and the DECs were defined as the target circRNAs.

Construction of ceRNA Regulatory Network

The target miRNAs that were predicted to combine with both the genes and the target circRNAs were selected for the establishment of the ceRNA network. Cytoscape was used to establish the mRNA–miRNA–circRNA regulatory network based on the mRNA–miRNA and miRNA–circRNA interactions.

Enrichment Analysis of Hub Genes and Predicted miRNAs

KEGG database is a database for systematic analysis of the metabolic pathways of gene products in cells and the functions of these genes in products. GO database is designed to establish a database to define and describe gene and protein functions applicable to various species. KEGG pathway and GO functional analyses were used to analyze DEGs using the “clusterProfiler” package in R. 17 And the results were visualized using the “GOplot” package (https://wencke.github.io/). The biological process and biological progress of predicted miRNAs were performed by FunRich software.

Animal Model of OP and RT‐qPCR Assays

According to the method in previous literature, the mouse model of OP was established. 18 , 19 This study was approved by the Animal Ethics Committee of Tianjin Hospital (Ethical Approval No: 2022 Medical Ethics Review 192).

C57BL/6 mice (11‐week‐old) were injected with botulinum toxin A (Botox) (2.5 unit/100 μl) in the right hindlimb (n = 8). Botox with a total dose of 2U/100g was injected into the quadriceps femoris and gastrocnemius. Control mice (n = 8) were injected with an equal volume of isotonic saline. The gait ability and the muscle thickness of the right hindlimb were monitored every day after injection until the mice were killed. Twenty‐one days after injection, the mice were killed by cervical dislocation. The right femurs of the mice were taken, and the muscle and soft tissue were removed. The femurs of the mice were imaged with micro‐CT, and the bone mineral density was observed.

The femurs of mice were mashed with liquid nitrogen in the mortar. Total bone tissue RNA was extracted with Trizol reagent and its purity measured. Reverse transcription of cDNA was performed using the kit according to the manufacturer's instructions. DNA was amplified by real‐time fluorescence polymerase chain reaction instrument and GAPDH was used as an internal control. Primer sequences are shown in Additional file 1: Table S1.

Evaluation of Candidate Drugs

As an easy‐to‐use intuitive enrichment analysis tool, the Enrichr platform (https://maayanlab.cloud/Enrichr/) is widely used to represent details of gene function. 20 The DSigDB database of the Enrichr platform was used to predict potential molecules for OP treatment. 21 And the drug that simultaneously target multiple target genes was singled out for further analysis.

Molecular Docking

The 3D structures of proteins encoded by hub genes and candidate drug were downloaded from Pubchem (https://pubchem.ncbi.nlm.nih.gov/) and PDB (http://www.rcsb.org/) databases. Molecular docking was predicted by autodock vina software, and the complex with the lowest binding energy was selected. 22 The interaction between the small molecule drug and the proteins was visualized.

Statistical Methods

The statistical software used in this study was RStudio and Graphpad Prism. For the analysis of bioinformatics data, three machine learning algorithms were adopted. PCR data were evaluated by independent sample t‐test.

Results

Flow Chart and Data Preprocessing

A protocol flow chart is shown in Figure 1 and baseline sample information in Table 1. PCA and cluster analysis showed that the OP and control samples formed different groupings (Figure 2(A, B)).

Fig. 1.

Fig. 1

Open in a new tab

Protocol flowchart

TABLE 1.

Clinical information of dataset

Data set Sample number Platforms
GSE7158
Control 14 GPL570
OP 12
GSE201543
Control 4 GPL20712
OP 6
GSE161361
Control 3 GPL28148
OP 3
Open in a new tab

Fig. 2.

Fig. 2

Open in a new tab

Data preprocessing. (A) PCA diagrams and (B) cluster analysis diagram of GSE7158

Identification of DEGs

A total of 179 DEGs were identified, of which 150 were up‐regulated and 29 were down‐regulated in OP samples compared with controls (Additional file 2: Table S2). A volcano map (Figure 3(A)) and a heat map (Figure 3(B)) of the 25 most strongly up‐ or down‐regulated genes are shown.

Fig. 3.

Fig. 3

Open in a new tab

DEGs of the integrated dataset. (A) Volcano map of the DEGs. (B) Heat map of part of the DEGs

Identification of Hub Genes

The results of RF showed that after five repetitions of 10‐fold cross‐validation, according to the principle of parsimony, when the number of genes reached 15, the error did not decrease significantly, and the first 15 genes extracted were defined as hub genes (Figure 4(A)). For a lambda value of lambda.min (0.04329) in the LASSO model, 27 non‐zero coefficient hub genes were apparent (Figure 4(B)). Ten hub genes were selected from the SVM‐RFE model, cross‐validation showed an error rate of misclassified diagnosis of 12.5%, which has the lowest error rate among all diagnosis models compared in (Figure 4(C)). BAG2, MME, SLC14A1, and TRIM44 were overlapping genes by the three algorithms including one down‐regulated (TRIM44) and three up‐regulated (BAG2, MME, and SLC14A1) genes, which were used as common hub genes for further study (Figure 4(D)). The ROC curves of BAG2, MME, SLC14A1, and TRIM44 revealed their probability as valuable diagnostic genes with AUCs of 0.952, 0.830, 0.949, and 0. 920, respectively (Figure 5(A–D)), indicating that the four hub genes had a high accuracy of predictive value.

Fig. 4.

Fig. 4

Open in a new tab

Screening hub genes. (A) Random forest (RF) algorithm to screen hub genes; (B) Least absolute shrinkage and selection operator (LASSO) regression algorithm to screen hub genes; (C) Based on support vector machine‐recursive feature elimination (SVM‐RFE) to screen hub genes. (D) Venn map of hub genes obtained by the three algorithms

Fig. 5.

Fig. 5

Open in a new tab

ROC curves for BAG2 (A), MME (B), SLC14A1 (C), and TRIM44 (D) were verified for GSE7158

Construction of the mRNA‐miRNA‐circRNA Network

Twenty‐seven miRNAs were predicted based on hub genes (Figure 6(A)). By intersecting the hub genes' predicted miRNAs and DEMs (Figure 6(B)) obtained from the GEO database, we confirmed three OP‐associated target miRNAs (hsa‐miR‐200b‐3p, hsa‐miR‐144‐3p, and hsa‐miR‐4306) finally (Figure 6(C)). A total of 14,731 circRNAs which could bind to the target miRNAs were obtained, and part of them were visualized (Figure 6(D)). Then by intersecting the predicted circRNAs and DECs (Figure 6(E)) obtained from GEO, 362 target circRNAs were confirmed (Figure 6(F)). A ceRNA regulatory network was established based on the mRNA–miRNA and miRNA–circRNA interactions (Figure 7).

Fig. 6.

Fig. 6

Open in a new tab

(A) Hub genes and predicted miRNAs. (B) Volcano map of differentially expressed microRNAs (DEMs) in GSE201543; (C) Venn map of DEMs and predicted miRNAs; (D) Target miRNAs and part of predicted circRNAs. (E) Volcano map of differentially expressed circRNAs (DECs) in GSE161361; (F) Venn map of DECs and predicted circRNAs

Fig. 7.

Fig. 7

Open in a new tab

CeRNA network. Interaction network among mRNA, microRNAs, and circRNAs

Function Enrichment Analysis of Hub Genes and Target miRNAs

DEGs were analyzed by GO functional enrichment and KEGG pathway analyses. For biological processes (BP), the hub genes were mainly enriched in negative regulation of protein polyubiquitination, respiratory system development, response to UV‐B, creatinine metabolic process, neuropeptide processing, urea transmembrane transport, and so on (Figure 8(A)). KEGG pathway analysis found that hub genes tended to be involved in protein processing in endoplasmic reticulum, Alzheimer's disease, hematopoietic cell lineageand protein digestion and absorption. (Figure 8(B)). FunRich software was used to perform enrichment analysis for predicted miRNAs. For BP, the target miRNAs were mainly enriched in outflow tract septum morphogenesis, positive regulation of canonical Wnt, receptor signaling pathway, response to hydrogen peroxide, regulation of cell growth, B cell homeostasis, neural tube closure, forebrain development, cellular response to transforming growth factor beta stimulus, phagocytosis, regulation of epidermal growth factor activated receptor activity, etc. (Figure 8(C)). For pathway analysis, the target miRNAs were mainly involved in ovarian steroidogenesis, pyrimidine metabolism, AGE‐RAGE signaling pathway in diabetic complications, leukocyte transendothelial migration, drug metabolism‐other enzymes, and oxytocin signaling pathway (Figure 8(D)).

Fig. 8.

Fig. 8

Open in a new tab

Enrichment analysis. (A) Enrichment GO BP annotation diagram of DEGs; (B) KEGG annotation diagram of DEGs; (C) Enrichment BP annotation diagram of the target miRNAs; (D) Pathway analysis annotation diagram of the target miRNAs

Verification of Biomarkers by RT‐qPCR

The OP model was identified by the naked eye and micro‐CT. By macroscopic observation, the muscles of the lower extremities of OP mice showed severe atrophy, and the micro‐CT result showed that the bone mass in OP model mice was lower than that in control mice (Figure 9(A, B)).

Fig. 9.

Fig. 9

Open in a new tab

Micro‐CT and RT‐qPCR. Micro‐CT images of OP (A) and control (B) mice. The expression of BAG2 (C), MME (D), SLC14A1 (E) and TRIM44 (F) in OP cell model by RT‐qPCR. *p‐value <0.05, **p‐value <0.01, ***p‐value <0.001

The femurs were extracted from OP model mice and controls and used for RT‐qPCR verification. Three genes (BAG2, MME, and SLC14A1) were up‐regulated and one (TRIM44) was down‐regulated in OP compared with control samples (Figure 9(C–F)). The gene expression levels were consistent with bioinformatics analysis results.

Candidate Drug Identification and Molecular Docking

Benzo[a]pyrene was predicted by the DSigDB database to be appropriate for candidates for OP treatment giving the highest combined scores (Table 2). Molecular docking abilities of benzo[a]pyrene with the four protein receptors identified above were predicted by autodock vina software. Lower binding energies indicate more stable products of ligand receptor interactions with a value of less than −5.0 kcal/mol indicating possible binding and of less than −7.0 kcal/mol indicating substantial likelihood of binding. Binding energies of three protein receptors (BAG2 [−9.3kcal/mol], MME [−11.1 kcal/mol], and SLC14A1 [−10.4kcal/mol]) with benzo[a]pyrene were less than −7.0 kcal/mol indicating that binding is substantially likely, binding energy of TRIM44 with benzo[a]pyrene was −6.7 kcal/mol indicating that binding is a possibility.

TABLE 2.

Suggested top drug compounds for osteoporosis

Index Name p‐value Odds ratio Combined score Genes
1 Benzo[a]pyrene CTD 00005488 0.002392 62,304 376056.1 SLC14A1, BAG2, MME, TRIM44
2 Trichostatin A HL60 DOWN 0.004257 35.96118 196.318 BAG2, TRIM44
3 Vorinostat HL60 DOWN 0.004667 34.26631 183.9175 BAG2, TRIM44
4 LY‐294002 MCF7 DOWN 0.004792 289.4638 1545.991 BAG2
5 Naringin HL60 DOWN 0.007778 175.0702 850.2272 BAG2
6 Hycanthone MCF7 DOWN 0.008175 166.3 799.3431 BAG2
7 LY‐294002 PC3 DOWN 0.008175 166.3 799.3431 BAG2
8 Ampyrone HL60 DOWN 0.00867 24.66881 117.1237 BAG2, TRIM44
9 Monorden MCF7 UP 0.009963 135.6939 625.3936 BAG2
10 Alvespimycin MCF7 UP 0.012343 108.9344 478.7281 BAG2
Open in a new tab

Interactions between Small Molecule Drug and Protein Receptors

A variety of interactions produced by docking show that benzo[a]pyrene and the four protein receptors can bind well (Figure 10(A–D)). Benzo[a]pyrene binds to BAG2 and forms six π‐π interactions with the amino acid residues TRP638 and TRP639 and four π‐cation interactions with LYS72. Benzo[a]pyrene forms five π‐π interactions with amino acid residues LEU204 and TYR314 of MME and two π‐anion interactions with the GLU205 residue of MME. In the complex formed with SLC14A1, benzo[a]pyrene forms three π‐σ interactions with the amino acid residue LEU121 and one π‐π interaction with TYR122. In the complex formed with TRIM44, benzo[a]pyrene forms one π‐S bond with amino acid residue CYS64 and three π‐cation interactions with LYS82.

Fig. 10.

Fig. 10

Open in a new tab

Molecular docking: proteinaceous receptors of BAG2 (A), MME (B), SLC14A1 (C) and TRIM44 (D) bind benzo[a]pyrene and the different forces and bonding modes produced by docking are represented by dotted lines of different colors

Discussion

In this study, BAG2, MME, TRIM44, and SLC14A1 were screened as the hub genes related to osteoporosis, and three target miRNAs, and 362 target circRNAs were predicted according to these genes, so as to construct a ceRNA network related to the pathogenesis of osteoporosis. In addition, we use the hub genes as target genes to predict the therapeutic drug molecules for osteoporosis and provide a basis for the clinical treatment of osteoporosis.

Role of Hub Genes

Bcl‐2 Associated Athanogene 2 protein encoded by BAG2 is a member of BAG family. BAG family has similar molecular structure and function, they compete with Hip for binding to the Hsc70/Hsp70 ATPase domain and promote substrate release. 23 , 24 Related studies have shown that BAG family proteins played a role in a variety of diseases, including cancer and neurodegenerative diseases. 25 , 26 In Parkinson's disease, BAG2 protected neurons through PINK1/PARKIN‐mediated pathway. 27 Yang et al. had shown that increased BAG2 expression in breast cancer blocked intracellular cathepsin B cleavage and led to tissue apoptosis and tumorigenesis. 28 We speculated that BAG2 could affect the occurrence of osteoporosis by affecting the apoptosis of osteoblasts and osteoclasts. Membrane metalloendopeptidase (MME) encodes a highly conservative zinc‐dependent membrane metalloproteinase (neprilysin, NEP), which is associated with many late‐onset hereditary diseases. 29 , 30 NEP is an important neuropeptidase and amyloid proteolytic enzyme, which plays a regulatory role in many organs, such as blood vessels, bones, and intestines. 31 Solute carrier family 14 member 1 (SLC14A1, also known as HUT11) encodes the primary human trabecular osteoblast precursors express human solute carrier family 14 (urea transporter), member 1, and it is a membrane transporter that mediates urea transport in red blood cells. 32 Previous studies illustrated shown that SLC14A1 played a role in the phenotypic transition from osteoblasts to adipocytes, but the exact mechanism was unclear. 33 Other studies showed that the expression of SLC14A1 decreased when bone marrow mesenchymal stem cells differentiated into cartilage. 34 We speculated that SLC14A1 acted on the nuclear hormone receptor peroxisome proliferator activated receptor‐γ. Because current studies showed SLC14A1 promoted the adipogenesis of BMSC and inhibited osteogenesis, 35 tripartite motif containing 44 (TRIM44) encodes a member of the tripartite motif (TRIM) family. 36 Unlike other typical TRIM family proteins, TRIM44 contains a zinc finger domain, which is usually found in ubiquitin‐specific proteases. 37 Previous studies demonstrated that TRIM44 was close related to the progression of malignant tumors. 38 , 39 Chen et al. showed that TRIM44 was overexpressed in multiple myeloma, resulting in increased bone destruction in mice. 40 TRIM44‐induced autophagy could be used as a cytoprotective response by degrading aggregates, but the role of osteoblasts and osteoclasts remained to be further studied. 41 To sum up, these hub genes play a potential role in the occurrence and development of osteoporosis and may be potential targets for the treatment.

CeRNA Regulatory Network

The ceRNA mechanism is a competitive interaction mechanism, which has been studied in many diseases and affects the prognosis of the disease; however, there are few studies on osteoporosis. 42 , 43 In our study, we constructed a ceRNA network to understand the role of ceRNA mechanism in osteoporosis, including two genes (SLC14A1 and TRIM44), three miRNAs (hsa‐miR‐200b‐3p, hsa‐miR‐144‐3p and hsa‐miR‐4306), and 362 circRNAs. As the core of ceRNA mechanism, miRNA plays an important role in osteoporosis. Chrdl1 was reported to be up‐regulated by inhibiting the expression of miR‐200b‐3p, thus inhibiting the apoptosis of MC3T3‐E1 cells induced by palmitic acid. 44 Another article reported that miR‐200b‐3p enhanced the growth and proliferation of osteoarthritis chondrocytes by inhibiting the expression of its target gene DNMT3A. 45 Also, some studies have shown that miR‐200b‐3p could inhibit the pyrolysis of bone marrow mesenchymal stem cells (BMSCs) by inhibiting the expression of Foxo3. 46 In terms of miR‐144‐3p, current studies reported that it inhibited osteogenic differentiation of BMSC by targeting BMPRB1. 47 However, another study showed that miR‐144‐3p also regulated osteoclast formation, proliferation, and apoptosis by regulating RUNK pathway. 48 Related studies reported that miR‐4306 played a regulatory role in the development of various tumors and affected the prognosis of tumors. 49 , 50 , 51 According to the current researches, there have been no studies to clarify the role of miR‐4306 in osteoporosis. This study suggested that miR‐4306 was differentially expressed in OP and normal samples, indicating that it played a potential role in the occurrence and development of OP, but further experiments were needed.

Potential Drug

Benzo[a]pyrene (BaP) is a typical polycyclic aromatic hydrocarbon, which usually exists in cigarette smoke and smog. Related studies had shown that polycyclic aromatic hydrocarbons were associated with cancer, immunosuppression, decreased resistance to viral and bacterial infections, and decreased endocrine function. 52 , 53 , 54 , 55 BaP could also inhibit osteoclast formation and bone resorption by binding to aryl hydrocarbon receptor (Ahr), which related to the inhibition of RANKL signal pathway. 56 , 57 Moreover, studies demonstrated BaP inhibited TGF‐β1/SMAD4 and TGF‐β1/ERK/AKT signaling pathways by activating Ahr, thus inhibiting the osteogenic differentiation of BMSC. 58 Thus, BaP and its antagonists have potential value in the treatment of osteoporosis. How to balance the effects of BaP and its antagonists will be a new direction in the treatment of osteoporosis.

Strengths and Limitations

The current study used bioinformatics approach to develop a new ceRNA network and potential drug for the treatment of OP. Firstly, scientific machine learning algorithms, such as RF, Lasso regression and SVM‐RFE algorithm, were used to identify hub genes, and these hub genes were verified by ROC curve and RT‐qPCR to make our results more convincing. Then miRNAs and circRNAs were predicted and the ceRNA network was constructed. Finally, potential therapeutic drugs were predicted based on hub genes. However, limitations still exist in this study. First, more in vivo studies are required to verify such issues. Also, a large number of prospective studies are needed to be further studied.

Conclusions

In general, in order to explore the molecular mechanism of osteoporosis pathogenesis, we screened hub genes and constructed a novel ceRNA regulatory network. In addition, we predicted potential drug molecules based on hub genes. Benzo[a]pyrene, as the drug molecule with the highest binding fraction to hub genes, may provide a new perspective for the optimal treatment of osteoporosis.

Conflict of Interest

No conflict of interest exists in the submission of this manuscript. We declare that all authors listed meet the authorship criteria according to the latest guidelines of the International Committee of Medical Journal Editors, and all authors are in agreement with the manuscript. All authors declared that the work has not been published previously, and is not under consideration for publication elsewhere, in whole or in part.

Authors Contributions

Conceptualization: Zheng Wang and Xiao‐fei Zhang. Data curation and formal analysis: Mao‐peng Wang, Shuo Yan, and Jie Zhao. Software and Methodology: Bing Li, Qing‐hang Qian, and Zheng‐xu Dai. Writing‐original draft: All authors. Writing—review and editing: Jun Liu and Xin‐long Ma. All authors read and approved the final manuscript.

Ethics Statement

Ethical approval was given for the use of data relating to human from public datasets.

Supporting information

Table S1. Primer sequences

Click here for additional data file. (15.9KB, docx)

Table S2. Differentially expressed genes

Click here for additional data file. (22.1KB, docx)

Table S3. Hub genes, corresponding miRNAs, and corresponding circRNAs

Click here for additional data file. (269KB, docx)

Acknowledgement

We acknowledge the GEO database for providing their platforms and their contributors for uploading meaningful datasets.

Zheng Wang and Xiao‐fei Zhang contributed equally to this work and share first authorship.

Contributor Information

Xin‐long Ma, Email: 16622080327@163.com.

Bing Li, Email: 1280337701@qq.com.

Jun Liu, Email: 13502117976@163.com.

Data Availability Statement

Publicly available datasets (GSE7158, GSE201543, and GSE161361) were analyzed in this study. All datasets were obtained from the GEO database. GSE7158 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE7158. GSE201543 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE201543. GSE161361 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE161361.

References

  • 1. Rachner TD, Khosla S, Hofbauer LC. Osteoporosis: now and the future. Lancet (London, England). 2011;377(9773):1276–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Eriksen EF. Cellular mechanisms of bone remodeling. Rev Endocr Metab Dis. 2010;11(4):219–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Shen G, Ren H, Shang Q, Zhao W, Zhang Z, Yu X, et al. Foxf1 knockdown promotes BMSC osteogenesis in part by activating the Wnt/β‐catenin signalling pathway and prevents ovariectomy‐induced bone loss. EBioMedicine. 2020;52:102626. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Liu Y, Han X, Bo J, Ma H. Wedelolactone enhances osteoblastogenesis but inhibits osteoclastogenesis through Sema3A/NRP1/PlexinA1 pathway. Front Pharmacol. 2016;7:375. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Catalano A, Loddo S, Bellone F, Pecora C, Lasco A, Morabito N. Pulsed electromagnetic fields modulate bone metabolism via RANKL/OPG and Wnt/β‐catenin pathways in women with postmenopausal osteoporosis: a pilot study. Bone. 2018;116:42–6. [DOI] [PubMed] [Google Scholar]
  • 6. Kanis JA, Norton N, Harvey NC, Jacobson T, Johansson H, Lorentzon M, et al. SCOPE 2021: a new scorecard for osteoporosis in Europe. Arch Osteoporos. 2021;16(1):82. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Atik OS, Gunal I, Korkusuz F. Burden of osteoporosis. Clin Orthop Relat R. 2006;443:19–24. [DOI] [PubMed] [Google Scholar]
  • 8. Center JR, Nguyen TV, Schneider D, Sambrook PN, Eisman JA. Mortality after all major types of osteoporotic fracture in men and women: an observational study. Lancet (London, England). 1999;353(9156):878–82. [DOI] [PubMed] [Google Scholar]
  • 9. Tella SH, Gallagher JC. Prevention and treatment of postmenopausal osteoporosis. J Steroid Biochem Mol Biol. 2014;142:155–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Palui R, Durgia H, Sahoo J, Naik D, Kamalanathan S. Timing of osteoporosis therapies following fracture: the current status. Ther Adv Endocrinol. 2022;13:1859485112. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Mao Z, Zhu Y, Hao W, Chu C, Su H. MicroRNA‐155 inhibition up‐regulates LEPR to inhibit osteoclast activation and bone resorption via activation of AMPK in alendronate‐treated osteoporotic mice. IUBMB Life. 2019;71(12):1916–28. [DOI] [PubMed] [Google Scholar]
  • 12. Ma S, Wang D, Ma C, Zhang Y. microRNA‐96 promotes osteoblast differentiation and bone formation in ankylosing spondylitis mice through activating the Wnt signaling pathway by binding to SOST. J Cell Biochem. 2019;120(9):15429–42. [DOI] [PubMed] [Google Scholar]
  • 13. Zhang M, Jia L, Zheng Y. circRNA expression profiles in human bone marrow stem cells undergoing osteoblast differentiation. Stem Cell Rev Rep. 2019;15(1):126–38. [DOI] [PubMed] [Google Scholar]
  • 14. Li Y, Bian M, Zhou Z, Wu X, Ge X, Xiao T, et al. Circular RNA SIPA1L1 regulates osteoblastic differentiation of stem cells from apical papilla via miR‐204‐5p/ALPL pathway. Stem Cell Res Ther. 2020;11(1):461. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. O'Brien J, Hayder H, Zayed Y, Peng C. Overview of MicroRNA biogenesis, mechanisms of actions, and circulation. Front Endocrinol. 2018;9:402. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Salmena L, Poliseno L, Tay Y, Kats L, Pandolfi PP. A ceRNA hypothesis: the Rosetta stone of a hidden RNA language? Cell. 2011;146(3):353–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Yu G, Wang L, Han Y, He Q. clusterProfiler: an R package for comparing biological themes among gene clusters. Omics J Integr Biol. 2012;16(5):284–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Warner SE, Sanford DA, Becker BA, Bain SD, Srinivasan S, Gross TS. Botox induced muscle paralysis rapidly degrades bone. Bone. 2006;38(2):257–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Poliachik SL, Bain SD, Threet D, Huber P, Gross TS. Transient muscle paralysis disrupts bone homeostasis by rapid degradation of bone morphology. Bone. 2010;46(1):18–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Chen EY, Tan CM, Kou Y, Duan Q, Wang Z, Meirelles GV, et al. Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinform. 2013;14:128. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Yoo M, Shin J, Kim J, Ryall KA, Lee K, Lee S, et al. DSigDB: drug signatures database for gene set analysis. Bioinformatics (Oxford, England). 2015;31(18):3069–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Trott O, Olson AJ. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. J Comput Chem. 2010;31(2):455–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Qin L, Guo J, Zheng Q, Zhang H. BAG2 structure, function and involvement in disease. Cell Mol Biol Lett. 2016;21:18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Takayama S, Sato T, Krajewski S, Kochel K, Irie S, Millan JA, et al. Cloning and functional analysis of BAG‐1: a novel Bcl‐2‐binding protein with anti‐cell death activity. Cell. 1995;80(2):279–84. [DOI] [PubMed] [Google Scholar]
  • 25. Papadakis ES, Reeves T, Robson NH, Maishman T, Packham G, Cutress RI. BAG‐1 as a biomarker in early breast cancer prognosis: a systematic review with meta‐analyses. Brit J Cancer. 2017;116(12):1585–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Wu J, Mamidi TKK, Zhang L, Hicks C. Unraveling the genomic‐epigenomic interaction landscape in triple negative and non‐triple negative breast cancer. Cancer. 2020;12(6):1559. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Qu D, Hage A, Don‐Carolis K, Huang E, Joselin A, Safarpour F, et al. BAG2 gene‐mediated regulation of PINK1 protein is critical for mitochondrial translocation of PARKIN and neuronal survival. J Biol Chem. 2015;290(51):30441–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Yang K, Bae E, Ahn SG, Pang K, Park Y, Park J, et al. Co‐chaperone BAG2 determines the pro‐oncogenic role of cathepsin B in triple‐negative breast cancer cells. Cell Rep. 2017;21(10):2952–64. [DOI] [PubMed] [Google Scholar]
  • 29. Higuchi Y, Hashiguchi A, Yuan J, Yoshimura A, Mitsui J, Ishiura H, et al. Mutations in MME cause an autosomal‐recessive Charcot‐Marie‐Tooth disease type 2. Ann Neurol. 2016;79(4):659–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Lupo V, Frasquet M, Sánchez‐Monteagudo A, Pelayo‐Negro AL, García‐Sobrino T, Sedano MJ, et al. Characterising the phenotype and mode of inheritance of patients with inherited peripheral neuropathies carrying MME mutations. J Med Genet. 2018;55(12):814–23. [DOI] [PubMed] [Google Scholar]
  • 31. Dutriez I, Salès N, Fournié‐Zaluski MC, Roques BP. Pre‐ and post‐natal ontogeny of neutral endopeptidase 24‐11 ('enkephalinase') studied by in vitro autoradiography in the rat. Experientia. 1992;48(3):290–300. [DOI] [PubMed] [Google Scholar]
  • 32. Olives B, Neau P, Bailly P, Hediger MA, Rousselet G, Cartron JP, et al. Cloning and functional expression of a urea transporter from human bone marrow cells. J Biol Chem. 1994;269(50):31649–52. [PubMed] [Google Scholar]
  • 33. Prichett WP, Patton AJ, Field JA, Brun KA, Emery JG, Tan KB, et al. Identification and cloning of a human urea transporter HUT11, which is downregulated during adipogenesis of explant cultures of human bone. J Cell Biochem. 2000;76(4):639–50. [PubMed] [Google Scholar]
  • 34. Schäfer R, Schnaidt M, Klaffschenkel RA, Siegel G, Schüle M, Rädlein MA, et al. Expression of blood group genes by mesenchymal stem cells. Brit J Haematol. 2011;153(4):520–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Nuttall ME, Gimble JM. Controlling the balance between osteoblastogenesis and adipogenesis and the consequent therapeutic implications. Curr Opin Pharmacol. 2004;4(3):290–4. [DOI] [PubMed] [Google Scholar]
  • 36. Reymond A, Meroni G, Fantozzi A, Merla G, Cairo S, Luzi L, et al. The tripartite motif family identifies cell compartments. EMBO J. 2001;20(9):2140–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Allen MD, Bycroft M. The solution structure of the ZnF UBP domain of USP33/VDU1. Protein Sci Publ Protein Soc. 2007;16(9):2072–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Peng R, Zhang P, Zhang C, Huang X, Ding Y, Deng B, et al. Elevated TRIM44 promotes intrahepatic cholangiocarcinoma progression by inducing cell EMT via MAPK signaling. Cancer Med US. 2018;7(3):796–808. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Kawabata H, Azuma K, Ikeda K, Sugitani I, Kinowaki K, Fujii T, et al. TRIM44 is a poor prognostic factor for breast cancer patients as a modulator of NF‐κB signaling. Int J Mol Sci. 2017;18(9):1931. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Chen Z, Lin T, Bi X, Lu G, Dawson BC, Miranda R, et al. TRIM44 promotes quiescent multiple myeloma cell occupancy and survival in the osteoblastic niche via HIF‐1α stabilization. Leukemia. 2019;33(2):469–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Lyu L, Chen Z, McCarty N. TRIM44 links the UPS to SQSTM1/p62‐dependent aggrephagy and removing misfolded proteins. Autophagy. 2022;18(4):783–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Qi X, Zhang D, Wu N, Xiao J, Wang X, Ma W. ceRNA in cancer: possible functions and clinical implications. J Med Genet. 2015;52(10):710–8. [DOI] [PubMed] [Google Scholar]
  • 43. Moreno‐García L, López‐Royo T, Calvo AC, Toivonen JM, de la Torre M, Moreno‐Martínez L, et al. Competing endogenous RNA networks as biomarkers in neurodegenerative diseases. Int J Mol Sci. 2020;21(24):9582. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Jing L, Hu B, Song QH. Lycium barbarum polysaccharide (LBP) inhibits palmitic acid (PA)‐induced MC3T3‐E1 cell apoptosis by regulating miR‐200b‐3p/Chrdl1/PPARγ. Food Nutr Res. 2020;64:64. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Wu J, Tao Y, Shang A, Wang W, Zhang Y, Hu L, et al. Effect of the interaction between MiR‐200b‐3p and DNMT3A on cartilage cells of osteoarthritis patients. J Cell Mol Med. 2017;21(10):2308–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Zhang L, Li S, Li J, Li Y. LncRNA ORLNC1 promotes bone marrow mesenchyml stem cell pyroptosis induced by advanced glycation end production by targeting miR‐200b‐3p/Foxo3 pathway. Stem Cell Rev Rep. 2021;17(6):2262–75. [DOI] [PubMed] [Google Scholar]
  • 47. Peng J, Zhan Y, Zong Y. METTL3‐mediated LINC00657 promotes osteogenic differentiation of mesenchymal stem cells via miR‐144‐3p/BMPR1B axis. Cell Tissue Res. 2022;388(2):301–12. [DOI] [PubMed] [Google Scholar]
  • 48. Wang C, He H, Wang L, Jiang Y, Xu Y. Reduced miR‐144‐3p expression in serum and bone mediates osteoporosis pathogenesis by targeting RANK. Biochem Cell Biol. 2018;96(5):627–35. [DOI] [PubMed] [Google Scholar]
  • 49. Liu L, He L, Li W, Zhao T, Li G, Xiu X, et al. The miR‐4306/IGF2R axis modulates the lung adenocarcinoma response to irradiation in vitro and in vivo. Transl Lung Cancer Res. 2021;10(12):4494–510. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Huang X, Pan J, Wang G, Huang T, Li C, Wang Y, et al. UNC5B‐AS1 promotes the proliferation, migration and EMT of hepatocellular carcinoma cells via regulating miR‐4306/KDM2A axis. Cell Cycle (Georgetown, Tex). 2021;20(20):2114–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Zhao Z, Li L, Du P, Ma L, Zhang W, Zheng L, et al. Transcriptional downregulation of miR‐4306 serves as a new therapeutic target for triple negative breast cancer. Theranostics. 2019;9(5):1401–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. de Oliveira Galvão MF, Sadiktsis I, Marques Pedro T, Dreij K. Determination of whole mixture‐based potency factors for cancer risk assessment of complex environmental mixtures by in vitro testing of standard reference materials. Environ Int. 2022;166:107345. [DOI] [PubMed] [Google Scholar]
  • 53. Bukowska B, Mokra K, Michałowicz J. Benzo[a]pyrene‐environmental occurrence, human exposure, and mechanisms of toxicity. Int J Mol Sci. 2022;23(11):6348. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Rurale G, Gentile I, Carbonero C, Persani L, Marelli F. Short‐term exposure effects of the environmental endocrine disruptor benzo(a)pyrene on thyroid axis function in zebrafish. Int J Mol Sci. 2022;23(10):5833. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55. Blanco GA, Cooper EL. Immune systems, geographic information systems (GIS), environment and health impacts. J Toxicol Environ Health B Crit Rev. 2004;7(6):465–80. [DOI] [PubMed] [Google Scholar]
  • 56. Safe S. Molecular biology of the Ah receptor and its role in carcinogenesis. Toxicol Lett. 2001;120(1–3):1–7. [DOI] [PubMed] [Google Scholar]
  • 57. Voronov I, Li K, Tenenbaum HC, Manolson MF. Benzo[a]pyrene inhibits osteoclastogenesis by affecting RANKL‐induced activation of NF‐kappaB. Biochem Pharmacol. 2008;75(10):2034–44. [DOI] [PubMed] [Google Scholar]
  • 58. Zhou Y, Jiang R, An L, Wang H, Cheng S, Qiong S, et al. Benzo[a]pyrene impedes self‐renewal and differentiation of mesenchymal stem cells and influences fracture healing. Sci Total Environ. 2017;587‐588:305–15. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table S1. Primer sequences

Click here for additional data file. (15.9KB, docx)

Table S2. Differentially expressed genes

Click here for additional data file. (22.1KB, docx)

Table S3. Hub genes, corresponding miRNAs, and corresponding circRNAs

Click here for additional data file. (269KB, docx)

Data Availability Statement

Publicly available datasets (GSE7158, GSE201543, and GSE161361) were analyzed in this study. All datasets were obtained from the GEO database. GSE7158 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE7158. GSE201543 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE201543. GSE161361 can be downloaded from: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE161361.

Articles from Orthopaedic Surgery are provided here courtesy of Wiley

RESOURCES