Introduction

Osteoporosis (OP) is a systemic metabolic skeletal disease characterized by osteopenia and disruption of bone microstructure, resulting in increased bone fragility and fracture incidence [1]. With the aging of the population and the widespread use of glucocorticoids, OP has become an “invisible killer” [2].With the global population over 60 years of age expected to reach two billion by 2050, the incidence of OP is expected to increase rapidly and have a major impact on older adults’ health and quality of life, especially postmenopausal females [3, 4]. In recent years, progress has been made in the research on the occurrence and pathogenesis of OP using molecular biology, cell biology, and other methods, mainly focusing on the Wnt/β-catenin [5, 6] and JAK/STAT signaling pathways [7, 8]. However, focusing only on a single gene or several genes from a local perspective can no longer satisfy this highly complex regulatory research. Based on the overall regulatory network, genes in OP are specifically expressed and closely related to regulatory factors [9], that are important in the occurrence and development of OP. Therefore, identify biomarkers for the occurrence and progression of OP is important for clinical diagnosis. Although the amount of biological data has increased exponentially with the rapid development of high-throughput technologies, most current informatics research has focused on studying differential gene expression, ignoring the possible correlations between genes and expression types. Weighted Gene Co-expression Network Analysis (WGCNA), a biotechnology based on scale-free networks, visually displays the interrelationships between various parts of biological systems and more accurately displays the characteristics of biological systems. This provides an important method for systematically studying diseases’ physiological and pathological processes [10]. WGCNA has been used in the genetic analysis of diseases like breast cancer [11], diabetes mellitus [12], Alzheimer’s disease [13], and coronary atherosclerotic heart disease [14], confirming that these modules and key genes play key roles in disease progression.

This study’s analysis of the GEO GSE56814 dataset revealed OP-associated hub transcriptional regulators. Bioinformatics analysis revealed the potential functions of OP-associated hub genes, and molecular biology experiments were performed to validate new biomarkers associated with OP progression (Fig. 1).

Fig. 1
figure 1

Complete flow diagram of the study

Full size image

Methods

Materials

Three normal and three osteoporotic bone tissues were collected from The Affiliated Hospital of Chengde Medical University (Chengde, Hebei, China) from 2020 to 2023, and are listed in Supplementary Table 1, which was approved by the Ethics Committee of The Affiliated Hospital of Chengde Medical University (2022.06.16/ No. CYFYLL2020240). The samples were immediately stored in -80℃ refrigerator for further experimental validation. The TRIzol reagent was procured from Invitrogen Co., Ltd. DEPC water was acquired from Soleibao Technology Co., Ltd. (Beijing, China). The HiFi-ScriptTM Reverse Transcription Kit was purchased from Kangwei Reagent Co., Ltd., and TB Green Premix Ex Taq II was obtained from Tiangen Biochemical Technology Co., Ltd. The antibodies SIRT1 and ZNF350, as well as GAPDH, were procured from Bioss Biological Technology Co., Ltd (Beijing, China, NO. BS-0921R and BS-1630R) and Gene Tex Biological Technology Co., Ltd (Beijing, China, NO. TEX112053 and GTX100118). Cy3-labeled Goat Anti-Rabbit IgG was obtained from Zhongshan Golden Bridge Biotechnology Co., Ltd. (Beijing, China). Chromatographic-grade solvents (chloroform, ethanol, xylene, and isopropanol) were acquired from Tianjin Yongda Chemical Reagent Co. Ltd. (Tianjin, China).

Datasets and samples

The GSE56814 dataset was downloaded from the GEO database using GeoQuery package [15]. In the GSE56814 dataset, hip bone mineral density (BMD) was measured using a Hologic 4500-W dual-energy X-ray absorptiometer (DXA) scanner (Hologic Corp., Waltham, MA), and the diagnostic criteria for high and low BMD were defined based on the top and bottom 30% of BMD values within our population. This dataset obtained total RNA from 60 milliliters of peripheral blood of 31 patients with OP and 42 healthy individuals. Subsequently, mRNA expression levels were quantified using the GeneChip Human Exon 1.0 ST Array (Affymetrix, Santa Clara, CA) in accordance with the manufacturer’s protocol. For WGCNA, we utilized annotation information from the GPL5175 biochip platform to match gene probes with their corresponding gene names.

Analyses of module stability and weighted co-expression networks

Based on the protocol of the WGCNA package in the R language, weighted co-expression networks (Healthy and OP) were constructed [16]. Paired Pearson correlations were applied to evaluate the weighted co-expression relationships among subjects in the adjacency matrix. An appropriate power value was determined if the degree of independence was 0.9. Gene modules were identified using a topological overlap matrix (TOM)-based hierarchical clustering approach based on dissimilarity measures (1-TOM). We then performed a 50-fold permutation test to classify modules not conserved in the normal and OP co-expression networks according to the module conservation function of the WGCNA package [17]. Modular genes are highly interconnected intra-module genes whose corresponding modules have the highest modular association (MM) values. We calculated the MM of each gene using the WGCNA function signedME, and considered the 100 genes with the highest MM values as key genes.

Construction of transcriptional regulatory factor network and functional enrichment analysis

The TRRUST Database (https://www.grnpediöttrust) [18, 19] records the regulatory relationships of transcription factors, which contain the targets corresponding to transcription factors and the regulatory relationships between transcription factors, describing a comprehensive transcriptional factor regulatory network. In this study, modular genes were mapped into the transcription factor regulatory network, and a modular gene-transcription factor regulatory relationship was obtained, which was visualized using the Cytoscape platform. We then used the ClusterProfiler package in R [20] to annotate genes associated with transcription factors and comprehensively studied the functional relevance of hub genes. Functional classification was assessed using Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG), and significance was set at P < 0.05 and q < 0.1.

Hub genes screening

After extracting the expression of each gene in the Transcriptional Regulatory Factor Network, we applied the Wilcoxon algorithm to screen hub differentially expressed genes in normal samples and OP samples. In order to visualize the results, we used the “ggpubr” package in R to draw violin plots. Statistical significance was set at P < 0.05.

Gene set variation analysis (GSVA)

GSVA is a nonparametric, unsupervised method for assessing gene pool enrichment that converts changes at the gene level to changes at the signaling pathway level. The biological functions of the samples were evaluated [21]. In this study, we obtained a set of genes from the Molecular Signature Database (version 7.0). We comprehensively evaluated the key genes using the GSVA package in R (http://www.bioconductor.org/gsva) to assess possible changes in biological functions between different samples.

Reverse-transcription quantitative polymerase chain reaction (RT-qPCR)

Total RNA of bone tissues obtained from the Chengde Medical College’s Affiliated Hospital in Chengde was isolated using a UNIQ-10 column TRIzol total RNA extraction kit, following the manufacturer’s instructions. cDNA was generated using the HiFi-ScriptTM Reverse Transcription Kit at 37℃ for 15 min, 85℃ for 5 s, and 4℃ in storage. PCR amplification was performed using TB Green Premix Ex Taq II. The primers used in this experiment were GAPDH, SIRT1, and ZNF350 (Homo sapiens; Supplementary Table 2). Each sample was tested in triplicate, and the relative expression of mRNA was calculated using the 2(−△△CT) method.

Immunofluorescence

Immunofluorescence analysis was performed as previously described [22]. Normal and surgical bone specimens were fixed in citric acid for 10 min aspirated, and infiltrated with 3% H2O2 for 20 min. Primary antibodies were applied overnight (SIRT1, 1:80; ZNF350, 1:120), followed by Cy3-conjugated goat anti-rabbit IgG for 20 min and DAPI for 20 min.

Statistical analysis

Data were analyzed using OriginPro8.0 software (USA), and expressed as mean ± standard error. One-way analysis of variance was used to analyze and compare the means of the different groups. P values < 0.05 were considered statistically significant.

Result

Identification of WGCNA modules

The original GSE56814 dataset was extracted from the GEO database, and R was used to perform background correction and normalization preprocessing on the original data. The Hclust function was used to remove batch effects, and the samples were randomly distributed to form a dendrogram. We selected power values 14 and 8 (the lowest power of the scale-free topology fitting index was 0.9) to generate OP and normal hierarchical clustering trees, respectively (Supplementary Fig. 2A-B). Normal samples were analyzed using WGCNA, and 15 modules were identified. The normal samples co-expression network modules were mapped to the disease sample network to evaluate the recurrence of module characteristics (Fig. 2A-B). To verify the stability of the identifier module, we used the module preservation function to calculate the preserved size of the module (Fig. 2C-D). The cyan module was retained with a lower degree of preservation (median rank = 2.0, Z-summary = 9.6), indicating that normal and OP states can be identified. The MM scores were calculated based on the WGCNA signed KME function, and the top 100 genes with the highest MM scores were considered key genes.

Fig. 2
figure 2

Clustering dendrograms and the characterization of gene modules identified by WGCNA. (A) OP samples; (B) normal samples; (C) module preservation median rank; and (D) Z-summary score

Full size image

Construction of transcriptional regulatory factor network and functional enrichment analysis

Key genes were imported into the TRRUST database to create transcriptional regulatory factors and key gene networks. The network consisted of 21 transcriptional regulatory factor, 48 key gene nodes and 79 interaction edges (Fig. 3). The top 10 biological processes were enriched, including type I interferon signaling pathway, transcription, DNA-templating, transcription from RNA polymerase II promoter, positive regulation of type I interferon production, positive regulation of transcription, and negative regulation of transcription (Fig. 4A). The transcription factor complex, nucleus, nucleoplasm, nuclear matrix, nuclear chromatin, and cytoplasm were related to the cellular component (Fig. 4B). The top 10 molecular functions, including transcription regulatory region DNA binding, transcription factor binding, protein binding, enzyme binding, DNA binding, and RNA polymerase II core promoter proximal region sequence-specific DNA binding, were identified (Fig. 4C). Regarding the KEGG pathway analysis (Fig. 4D), it was apparent that the targets mostly participated in the Toll-like receptor, TNF, PI3K-Akt, Osteoclast differentiation, JAK-STAT, and Chemokine signaling pathway.

Fig. 3
figure 3

Transcriptional regulatory factor -key genes network. Orange nodes represent transcriptional regulatory factor, and purple nodes represent key genes

Full size image
Fig. 4
figure 4

Functional enrichment analysis of targets identified in the transcriptional regulatory factor-gene network. (A) Biological process analysis; (B) cellular component analysis; (C) molecular function analysis; and (D) KEGG pathway analysis

Full size image

Hub transcriptional regulatory factors screening and analysis

The plot in Fig. 5 shows the expression of differentially expressed transcriptional regulatory factors in normal and OP samples based on Wilcoxon algorithm, namely SIRT1, and ZNF350. A panel of 50 hallmark genes was screened in the normal and OP samples using GSVA to identify important signaling pathways (Fig. 6A); 31 significantly expressed hallmark gene sets were identified by GSVA in the samples with high expression of SIRT1. IL6 Jak Stat3 signaling, apoptosis, and PI3K-Akt-mTOR signaling were significantly differentially expressed in different expression groups of SIRT1. Additionally, PI3K-Akt-mTOR, oxidative phosphorylation, P53 pathway, apoptosis, and hypoxia were significantly differentially expressed in the different expression groups of ZNF350 (Fig. 6B), suggesting SIRT1 and ZNF350 promoted the progression of OP by affecting oxidative phosphorylation, apoptosis, PI3K-Akt-mTOR signaling, and p53 pathway.

Fig. 5
figure 5

The expression levels of the hub differentially expressed genes in normal and OP samples. (A) SIRT1 and (B) ZNF350

Full size image
Fig. 6
figure 6

GSVA analysis showing the activated hallmark pathways. (A) SIRT1 and (B) ZNF350

Full size image

Experimental validation

Hub genes in normal (n = 3) and OP (n = 3) samples were verified using RT-qPCR. The hub gene expression levels (SIRT1 and ZNF350) in normal and OP samples were as follows: 0.96 ± 0.05, 2.13 ± 0.19; 0.99 ± 0.90, 3.03 ± 0.54, respectively, indicating that the mRNA expression of hub gene detected in OP samples was significantly or extremely significantly increased compared with normal samples (P < 0.05 or, P < 0.01) (Fig. 7). Furthermore, we confirmed the differential expression of hub genes in the normal and OP samples using immunofluorescence. The immunostaining intensities of SIRT1 and ZNF350 were higher in the OP samples than in the normal samples (Fig. 8).

Fig. 7
figure 7

Hub gene mRNA expression levels in normal and OP samples. (A) SIRT1 and (B) ZNF350. Data are presented as the mean ± SD (n = 3). * P < 0.05 and ** P < 0.01

Full size image
Fig. 8
figure 8

Immunofluorescence staining of SIRT1 and ZNF350. (A) SIRT1 and (B) ZNF350. The red immunofluorescence in the pictures represent SIRT1 or ZNF350. Cell nuclei were counterstained with DAPI

Full size image

Discussion

OP, a systemic bone metabolic disease with a complex pathogenesis, affects the quality of life and causes serious clinical consequences and social burden; however its specific pathogenesis has not yet been fully clarified [23]. This study, employed bioinformatics analysis to screen OP transcriptional regulatory biomarkers and validate the molecular biology experiments. WGCNA was used to calculate the co-expression relationships between molecules in a network, identify hub genes in modules, and gain insight into the expression profiles associated with hub genes and functional gene connections. This study used GSE56814 expression profiles comprising 73 blood samples from 31 patients with OP and 42 healthy controls. The WGCNA algorithm was used to filter the key modules and calculate the conservative degree of the modules (Fig. 2), where the lower preservation cyan module can distinguish normal and OP networks based on normal and disease samples; therefor, the cyan module was selected as the key module for the following analysis. The top 100 genes with the highest MM scores were mapped to the TRRUST database to construct the transcriptional factor regulatory network. Targets involved in the transcriptional factor-gene regulatory network were enriched in the Toll-like receptor, TNF, PI3K-Akt, JAK-STAT and osteoclast differentiation signaling pathway. The PI3K-Akt signaling pathway is critical in differentiating skeletal cells such as osteoblasts, chondrocytes, myoblasts and adipocytes [24,25,26]. Xi et al. studied a rat OP model and a cultured osteoblast model. They showed that the PI3K/Akt signaling pathway suppresses OP by promoting osteoblast proliferation and differentiation and enhancing bone formation [27]. Osteoclast differentiation and activation have been elucidated through relevant biological assays of a family of proteins known as the tumor necrosis factor receptor (TNFR)/TNF-like proteins. This family includes osteoprotegerin, receptor activator of nuclear factor (NF)-κB (RANK), and the RANK ligand (RANKL) [28]. Cao et al. investigated the underlying mechanism of TRPV4 expression in osteoclast differentiation using M-CSF and RANKL-induced RAW264.7 macrophage cells [29]. They found that the downregulation of TRPV4 probably inhibits autophagy in osteoclast differentiation, thereby inhibiting OP. Yin et al. believed that glycyrrhizic acid (Gly) could inhibit NF-κB, ERK, and JUK signaling pathways in vitro, strongly inhibit osteoclast differentiation and bone resorption, and has a bone-protective effect on OVX mice [30].The JAK-STAT signaling pathway was primarily recognized as a receptor-activated pathway in response to interferon (IFN)-γ and interleukin 6 (IL-6) cytokine family [31], and it is critical for cytokine effects on osteoblast proliferation and differentiation. Xu et al. suggested that fibroblast growth factor 23 (FGF23) is involved in bone and cartilage metabolism through the JAK/STAT pathway [7].

We applied the Wilcoxon algorithm to screen the hub differentially expressed transcriptional regulatory factors SIRT1 and ZNF350. Furthermore, RT-PCR and immunofluorescence assay demonstrated that the mRNA and protein expression levels of SIRT1 and ZNF350 differed significantly between normal and OP samples in clinical studies (n = 3), suggesting that SIRT1 and ZNF350 could considered as biomarkers for diagnosing OP. SIRT1(Silent mating type information regulation 2 homolog- 1) is a type III acetyltransferase important in the pathophysiology of metabolic diseases, degenerative diseases, cancer, and aging [32]. Studies have shown that SIRT1 expression increases in the nucleus and cytoplasm of peripheral blood mononuclear cells in patients with OP [33]. Similarly, related studies have confirmed that the specific knockout of SIRT1 in preosteoclasts has no significant effect on the volume of cancellous bone; in contrast, the specific knockout of SIRT1 in mature osteoblasts results in cancellous bone. The volume of the body is significantly reduced, and the activation or overexpression of SIRT1 reduces bone loss due to age [34,35,36].

SIRT1, with diverse roles in bone biology, is central to various cellular processes. Louvet et al. linked SIRT1 to bone mass in OP. In the separation-based anorexia (SBA) mouse model, activating SIRT1 reduced adipogenesis of BMSCs and increased osteogenesis, while inhibition had opposite effects. Resveratrol restored SIRT1 levels, which normalized bone parameters. These findings, seen through acetylation levels of transcription factors Runx2 and Foxo1 [37], underscore SIRT1’s intricate interplay in bone health.

Moreover, SIRT1, a prominent cellular regulator, is a sensor of cellular energy and metabolism [38]. Evidence supports its involvement in endocrine and metabolic disorders, notably OP [39]. Observations indicate SIRT1 influences bone remodeling through intricate signaling pathways, including the somatotropic axis [40]. The potential association with steroid hormone signaling is substantiated by upregulated SIRT1 expression upon estrogen treatment. Conversely, ovariectomy (OVX) induces a decline in SIRT1 expression [41].

SIRT1 plays a pivotal role in epigenetic regulation of bone physiology. Nutrients impact metabolic processes [42], contributing to SIRT1’s role in orchestrating tissue homeostasis through protein deacetylation [43]. Research suggests age-related pathologies, including OP, could be mitigated by modulating adipose mobilization, myogenic differentiation, dietary intake, and overall metabolism [44]. This interconnected understanding emphasizes SIRT1’s multifaceted impact on bone health and its potential therapeutic relevance.

ZNF350, also known as zinc-finger 350, participates in cell proliferation during the development of various diseases, such as breast, colon, and cervical cancers [45,46,47]. ZNF350 also acts as a transcriptional corepressor, inhibits the expression of SNAI2, and regulates epithelial mediator transformation (EMT)-associated genes such as MMP9, KAP1, and SNAI2, thereby inhibiting tumor EMT [48].

Research suggests that ZNF350 plays a role in the pathogenesis of diseases through involvement in ferroptosis. ZNF350 forms a transcription factor complex by binding with IRF1. It directly associates with the GPX4 promoter region, suppressing GPX4 transcription and inducing ferroptosis [49]. Moreover, ZNF350, by regulating NCOA4 transcription, modulates iron accumulation, lipid peroxidation, and ferroptosis, thereby influencing glioma progression [50]. Additionally, ZNF350 typically functions as a monomer, impacting the transcription of downstream genes. For example, ZNF350 impedes cervical cancer progression by directly binding to the MMP9 promoter region and inhibiting its transcription [47]. Nevertheless, the role of ZNF350 in bone homeostasis has not been thoroughly investigated.

Our investigation has identified and substantiated ZNF350 as a diagnostic biomarker for OP. The prospective trajectory of our research will center on meticulous validation across diverse sample cohorts and the implementation of a multi-method approach. Additionally, our focus will extend to elucidating the intricate mechanistic aspects underpinning the diagnostic efficacy of ZNF350. This holistic strategy is poised to deepen our comprehension of OP and pave the way for the development of more refined and effective diagnostic methodologies. Nevertheless, this study’s limitations include its reliance on the small sample size of the GSE56814 dataset and clinical verification research, which such constraints may introduce bias and restrict the representation of the target population. Thus, it is imperative to conduct further clinical experiments to validate these findings in the future, and our team remains committed to monitoring this research closely.

Conclusion

We identified key transcriptional regulators and pathways associated with OP occurrence and development through a series of comprehensive bioinformatics analyses. The identified transcriptional regulators included SIRT1 and ZNF350, which can serve as potential biomarkers for the diagnosis and prognosis of OP and will be further validated in further studies.