The GSE35958 dataset (19) included mRNA expression profiles of mesenchymal stem cells (MSCs) from bone marrow in 5 primary patients with osteoporosis and 4 age-matched, nonosteoporotic donors based on the GPL570 platform. The GSE56815 dataset (20) included mRNA expression profiles of blood monocytes from 40 females with low hip BMD and 40 females with high BMD based on the GPL96 platform. GSE159121 included miRNA and circRNA expression profiles of peripheral blood samples from 3 healthy controls and 3 patients with osteoporosis based on the GPL16791 platform.

Differential expression analysis for mRNAs between osteoporosis and controls in GSE35958 and GSE56815 was performed using the limma R package (21). Differential expression analysis for miRNAs and circRNA between osteoporosis and controls in GSE159121 was performed using the DEseq2 R package (22). The genes with a P-value <0.05 when comparing between patients with osteoporosis and controls were assigned as differentially expressed mRNAs, miRNAs, and circRNAs (DEmRs, DEmiRs, and DEcircRs).

The enrichment analysis of Gene Ontology (GO) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways for DEmRs were performed using the clusterProfiler R package (23). A P-value <0.05 was set to screen the significant enrichment results.

The target circRNAs of DEmiRs were predicted using the circBank database (http://www.circbank.cn/) and miRanda database (http://www.microrna.org/microrna/home.do). The target mRNAs of DEmiRs were predicted using the miRanda database. According to the predicted negative regulation of miRNAs, a network of competing endogenous RNAs (ceRNAs) was constructed.

A nomogram was built using logistic regression analysis with the rms R package (v3.5.3) (24) based on DEmRs to evaluate the risk assessment of osteoporosis. Forest plots were also constructed using logistic regression analysis with the forestplot R package (v4.0.3) (25). Receiver operating characteristic (ROC) curve analysis was performed using the pROC package (v1.16.2) (26).

Bone marrow tissue from 5 patients (2 males) with osteoporosis and 5 patients (2 males) with accidental fracture (without osteoporosis) were collected using surgery from August 2020 to November 2021 in the Sixth Affiliated Hospital of Xinjiang Medicine University (Urumqi, China). The age of the osteoporotic patients ranged from 31 to 96 years, with a mean age of 72.34±12.47 years, and that of patients without osteoporosis ranged from 31 to 95 years, with a mean age of 70.50±12.34 years. All participants had normal levels of parathyroid hormone with a mean level of 48.05±12.76 (normal range, 9–90 pg/ml), thyroid stimulating hormone with a mean level of 2.31±0.69 (normal range, 0.3-4.5 mIU/l). Inclusion criteria for osteoporosis were as follows: T-score of <-2.5 as detected by X-ray absorptiometry. Exclusion criteria were as follows: i) Patients with a history of brittle fractures within the past 12 months; ii) patients with bone diseases; iii) patients with a previous history of external bone irradiation or implanted radiation therapy; and iv) patients who had suffered from any type of cancer within the past 5 years. The present study was approved by the Ethics Committee of the Sixth Affiliated Hospital of Xinjiang Medicine University (approval no. LFYLLSC20191231-06). Patients were informed about the study and provided written informed consent.

Primary human bone marrow mesenchymal stem cells (BMSCs) were obtained from Pricella (https://www.procell.com.cn/view/2301.html). The use of BMSCs was approved by the Ethics Committee of the Sixth Affiliated Hospital of Xinjiang Medicine University (approval no. LFYLLSC20191231-06). BMSCs were cultured in DMEM medium (Hyclone; Cytiva) containing 10% FBS (Beijing Solarbio Science & Technology Co., Ltd.) and 100 U/ml penicillin and streptomycin at 37°C with 5% CO₂. The recombinant plasmid pcDNA3/circ_0070304 for circ_0070304 expression and the negative control plasmid (NC) were purchased from Shanghai Gemma Medical Technоlоgy Cо., Ltd. These plasmids (500 ng/ml) were transfected with Lipofectamine 2000 (Invitrogen; Thermo Fisher Scientific, Inc.) into BMSCs at 37°C for 48 h, then immediately used for subsequent experiments.

For osteogenic differentiation, BMSCs were cultured in osteogenic induction medium (50 mM ascorbic acid, 100 nM dexamethasone, 10 mM β-glycerophosphate, and 15% FBS) for 2 weeks.

Subsequently, BMSCs (3×10⁵/well) were fixed in 60% isopropanol for 60 sec at 37°C and washed twice with PBS. These cells were then stained with 1% Alizarin Red (Beijing Solarbio Science & Technology Co., Ltd.) staining for 3 min at 37°C. After washing the cells with PBS twice, the cells were observed and imaged using a fluorescence inverted microscope (Leica Microsystems GmbH). The type of microscope was a fluorescence inverted microscope, however fluorescence was not used.

For the dual luciferase reporter assay, circ_0070304 and the ring finger and CCCH-type domains 2 (RC3H2) 3′ untranslated region (3′UTR) were synthesized and cloned into psiCHECK-2 vectors, termed circRNA-wild-type (WT) or RC3H2-WT. The mutant (Mut) sequences of circ_0070304 and RC3H2 3′UTR were also synthesized and cloned into the same vectors to obtain circRNA-Mut or RC3H2-Mut. Then, these vectors (Shanghai Gemma Medical Technоlоgy Cо., Ltd.) were cotransfected with miR-183-5p mimics (forward, 5′-UAUGGCACUGGUAGAAUUCACU-3′ and reverse, 5′-UGAAUUCUACCAGUGCCAUAUU-3′) or miR-183-5p control (forward, 5′-UUCUCCGAACGUGUCACGUTT-3′ and reverse, 5′-ACGUGACACGUUCGGAGAATT-3′) into BMSCs using Lipofectamine 2000 (Invitrogen; Thermo Fisher Scientific, Inc.). The sequences were designed by Shanghai Gemma Medical Technоlоgy Cо., Ltd., according to the sequence in the miRbase database. To detect the fluorescence in each well, the Dual Luciferase Reporter Assay System (Thermo Fisher Scientific, Inc.) was applied after 36 h of transfection. Relative Luciferase activity is the ratio of Renilla luciferase activity normalized to firefly fluorescence activity.

Total RNA was extracted from bone marrow tissue or BMSCs using TRIzol (Invitrogen; Thermo Fisher Scientific, Inc.). Reverse transcription of RNA into cDNA was performed using Superscript II reverse transcriptase (Invitrogen; Thermo Fisher Scientific, Inc.) according to the maufacturer's instructions. RT-qPCR was then performed with a SYBR Green PCR kit (Invitrogen; Thermo Fisher Scientific, Inc.). The reaction conditions were as follows: 95°C for 10 min; 95°C for 10 sec, 40 cycles; 60°C for 15 sec; 72°C for 20 sec; 72°C for 10 min. GAPDH and U6 were used as internal reference genes to normalize the expression of target genes using the 2^−ΔΔCq method (27). Primer sequences were designed by Primer Premier 5 software (Premier Biosoft International) and are shown in Table SI.

Proteins were extracted from bone marrow tissue or BMSCs with RIPA lysis buffer (Beyotime Institute of Biotechnology) on ice. Quantification was performed using BCA Protein Assay Kit (Beyotime Institute of Biotechnology). A total of 50 µg proteins were segregated using 10% SDS-PAGE and then transferred onto PVDF membranes (MilliporeSigma). After blocking with 5% nonfat dry milk for 2 h at 37°C, the membranes were incubated with primary antibodies against ZFP36L1 (1:1,000; cat. no. ab230507; Abcam), P4HB (1:5,000; cat. no. ab137119; Abcam), RC3H2 (1:1,000; cat. no. ab241406; Abcam), TRIM28 (1:5,000; cat. no. ab109287; Abcam), WNK1 (1:2,500; cat. no. ab181029; Abcam), GAPDH (1:5,000; cat. no. ab181602; Abcam) overnight at 4°C. The membranes were then incubated with HRP-conjugated secondary antibody (anti-rabbit IgG; 1:2,500; cat. no. ab6734; Abcam) for 2 h at room temperature. GAPDH was used as an internal reference protein. Finally, ECL chemiluminescent liquid (Beijing Solarbio Science & Technology Co., Ltd.) was employed to identify the protein bands in the chemiluminescence using ChemiDoc MP imaging system (Bio-Rad Laboratories, Inc.). Densitometric analysis was performed using Image J software (v3.0; National Instiutes of Health).

Statistical significance between groups was analyzed by unpaired two-tailed Student's t-test using SPSS 20.0 software (IBM Corp.). At least three independent experiments were performed for all assays. All data are shown as the mean ± standard deviation (SD). Correlation analysis was performed using Pearson's correlation. P<0.05 was considered to indicate a statistically significant difference.

To identify mRNAs potentially associated with osteoporosis, differential analysis of mRNA expression was performed. In GSE35958, a total of 2,508 DEmRs were identified in patients with osteoporosis compared with controls (Fig. 1A). In GSE56815, 550 DEmRs (Fig. 1B) were identified. Finally, through intersection analysis, 110 DEmRs that were present in both GSE35958 and GSE56815 (Fig. 1C) were identified. There was a clear difference in the expression of these intersectional DEmRs between osteoporotic patients and controls (Fig. 1D). These DEmRs may be associated with osteoporosis.

To explore the potential pathological mechanism of osteoporosis, biofunction enrichment analysis of intersectional DEmRs was performed. The results of GO analysis for intersectional DEmRs included biological processes (BP), cellular component (CC), and molecular function (MF) (Fig. 2A). In the BP category, it was identified that ‘positive regulation of nucleic acid-templated transcription’, ‘negative regulation of transcription, DNA-templated transcription’, and ‘response to estradiol’ were mainly enriched. For CC, ‘intracellular membrane-bound organelle’, ‘nucleus’, and ‘focal adhesion’ were significantly enriched. ‘RNA binding’, ‘cadherin binding’, and ‘mRNA binding’ were significantly enriched in MF. In addition, the ‘estrogen signaling pathway’, ‘thyroid hormone signaling pathway’, and ‘adherens junction’ were significantly enriched in KEGG pathways (Fig. 2B).

To identify the regulatory network associated with osteoporosis, 33 DEmiRs (Fig. 3A) and 228 DEcircRs (Fig. 3B) in GSE159121 were identified. The circRNAs targeted by the DEmiRs were then predicted using the circBank database and miRanda database, of which four overlapped with DEcircRs: circ_0065265, circ_0002632, circ_0005812, and circ_0070304 (Fig. 3C). Next, the mRNAs targeted by the DEmiRs were predicted using the miRanda database, and 27 downstream mRNAs overlapped with DEmRs (Fig. 3D). The circRNAs, miRNAs and mRNAs were synthesized into ceRNA networks (Fig. 3E). It was then determined that circRNAs were all downregulated in the network. According to the negatively regulated sponge relationship, the targeted DEcircRs and DEmRs that are negatively regulated by miRNAs were identified. Finally, miR-183-5p (upregulation) was identified as a key DEmiR, circ_0002632 and circ_0070304 (downregulation) were identified as key DEcircRs, and ZFP36L1, P4HB, RC3H2, TRIM28, and WNK1 (downregulation) were identified as key DEmRs.

To evaluate the key DEmRs in the ceRNA network, a nomogram (Fig. 4A) was first constructed. P4HB, ZFP36L1, and RC3H2 contributed the most to the total score. The predictive values were consistent with the observed values according to the calibration curve (Fig. 4B). Furthermore, the distribution of key DEmRs was visualized using a forest plot (Fig. 4C). RC3H2 was a protective factor for osteoporosis. The area under the ROC (AUC) value of RC3H2 was 0.68, which had predictive diagnostic power (Fig. 4D). Therefore, it was identified that circ_0002632 and circ_0070304 may compete with RC3H2 for binding to miR-183-5p and may serve as promising targets for the diagnosis and treatment of osteoporosis.

To verify the aberrant expression of key genes, bone marrow tissue from patients was collected. In the RT-qPCR results, key DEmRs were all downregulated in osteoporosis compared to controls, circ_0070304 was significantly downregulated, and miR-183-5p was upregulated (Fig. 5A). According to the results of WB, the protein levels of ZFP36L1, P4HB, RC3H2, TRIM28, and WNK1 were all downregulated in osteoporosis (Fig. 5B).

Dual luciferase reporter assays were then performed to determine the specific 3′UTR binding between genes within the ceRNA network (circ_0070304/miR-183-5p/RC3H2). Notably, circRNA-WT or circRNA-Mut was cotransfected with miR-183-5p mimics or miR-183-5p control into hBMSCs. The luciferase activity of circRNA-WT was significantly inhibited by miR-183-5p mimics, while the luciferase activity of circRNA-Mut was not significantly changed by miR-183-5p mimics (Fig. 5C). The luciferase activity of RC3H2-WT was significantly inhibited by miR-183-5p mimics, and that of RC3H2-Mut was not significantly changed by miR-183-5p mimics (Fig. 5D). The results of correlation analysis showed a significant negative correlation between circ_0070304 and miR-183-5p with a correlation coefficient of −0.85 or between miR-183-5p and RC3H2 with a correlation coefficient of −0.77 (Fig. 5E).

To elucidate the function of circ_0070304 in osteoporosis, the plasmid pcDNA3/circ_0070304 was transfected into BMSCs to overexpress circ_0070304. The expression of circ_0070304 was significantly increased in the overexpression group compared to the NC group (Fig. 6A). In the overexpression group, the RNA level of miR-183-5p was reduced (Fig. 6A). WB demonstrated that RC3H2 protein expression was significantly increased in the circ_0070304 overexpression group compared to the NC group (Fig. 6B).

Subsequently, osteogenic induction was performed and BMSCs transfected with plasmid pcDNA3/circ_0070304 and NC were stained with Alizarin Red. Increased calcium deposition in the circ_0070304 overexpression group was revealed compared to the NC group (Fig. 6C). These results all suggested that circ_0070304 promotes osteoblastic differentiation.