Identification of biomarkers associated with endoplasmic reticulum stress-related cell death in osteoporosis based on bulk and single-cell transcriptomic analyses and experimental validation

Abstract

Osteoporosis (OP) is a metabolic bone disease characterized by low bone mineral density (BMD), and its pathogenesis involves endoplasmic reticulum (ER) stress-related cell death. This study aimed to identify diagnostic biomarkers associated with ER stress-related cell death in OP and explore their underlying mechanisms. The training dataset (GSE56815), validation dataset (GSE56814), and single-cell RNA sequencing (scRNA-seq) dataset (GSE147287) were downloaded. Differentially expressed genes (DEGs) between OP patients and controls were identified. Candidate genes were obtained by intersecting DEGs with ER stress-related genes and programmed cell death (PCD)-related genes. Machine learning was used to screen intersection genes, and biomarkers were determined via expression level analysis. Gene set enrichment analysis (GSEA), immune cell infiltration analysis, drug prediction and molecular docking, scRNA-seq analysis, key cell screening, cell communication analysis, and pseudotime analysis were performed. Finally, reverse transcription quantitative polymerase chain reaction (RT-qPCR) were further conducted. A total of 28 candidate genes were obtained by intersection. CAMKK2 and DAPK3 were confirmed as biomarkers, and were consistently down-regulated in both datasets and verified by RT-qPCR. GSEA analysis revealed that biomarkers were enriched in cytokine-cytokine receptor interaction. Correlations between biomarkers and activated dendritic cells were found via immune cell infiltration analysis. Preliminary computational analyses indicated that drugs including calcitriol and danazol may potentially interact with the biomarkers in a stable manner. Bone marrow-derived mesenchymal stem cells (BM-MSCs) were identified as potential key cells via scRNA-seq analysis. Complex interactions involving BM-MSCs, such as ANGPTL4-CDH11 mediating BM-MSC self-communication, were revealed by cell communication analysis. Dynamic expression of biomarkers during BM-MSC differentiation was shown by pseudotime analysis: CAMKK2 fluctuated with differentiation stages, while DAPK3 shifted from high to low then high expression. CAMKK2 and DAPK3 were confirmed as diagnostic biomarkers for OP, providing insights into OP diagnosis and potential therapeutic targets.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-43744-w.

Introduction

Osteoporosis (OP) is a systemic skeletal disease characterized by diminished bone mass and microarchitectural degradation of bone tissue, which collectively result in enhanced bone fragility and a consequent increase in susceptibility to fractures¹. Osteoporosis affects an estimated 200 million individuals globally and is particularly common in postmenopausal women. With the ongoing aging of the global population, both the incidence of the disease and the associated healthcare burden are steadily increasing^1,2. Current therapeutic strategies include anti-resorptive agents (e.g., bisphosphonates, denosumab), bone-forming agents (e.g., teriparatide), and dual-action drugs (e.g., romosozumab), often requiring sequential regimens to sustain therapeutic efficacy¹. Nevertheless, long-term pharmacotherapy for osteoporosis poses significant clinical challenges: bisphosphonates are associated with risks of osteonecrosis of the jaw and atypical femoral fractures; denosumab discontinuation can trigger rebound increases in bone turnover and elevated fracture risk; and bone-forming agents provide only transient anabolic effects, requiring subsequent anti-resorptive therapy to maintain skeletal benefits^1,3,4. Furthermore, available therapies exhibit limited efficacy in improving bone quality, and a subset of patients respond inadequately to existing options^5,6. These constraints underscore the need to elucidate the molecular mechanisms governing bone metabolism, which is imperative for developing novel therapeutic approaches.

Growing evidence indicates that endoplasmic reticulum (ER) stress-mediated cell death plays an increasingly critical role in osteoporosis pathogenesis. ER stress, characterized as an adaptive cellular response to the accumulation of misfolded or unfolded proteins, may initiate key pathological processes when dysregulated. However, persistent or excessive ERS can trigger cell death through signaling pathways such as PERK–eIF2α–ATF4–CHOP and IRE1α–XBP1^7,8. In the context of osteoporosis (OP), cadmium exposure has been shown to induce ERS via reactive oxygen species (ROS), activating the PERK pathway while suppressing the Nrf2/NQO1 axis, ultimately leading to osteoblast apoptosis⁷. Similarly, high cholesterol levels promote osteoblast apoptosis through ERS activation⁹. Loss of COPB1 induces both ERS and ferroptosis by repressing SLC7A11 transcription via ATF6, thereby impairing cystine uptake and exacerbating OP progression¹⁰. Furthermore, bisphosphonates can trigger ERS-mediated apoptosis in lymphatic endothelial cells via the NAD⁺/SIRT6/XBP1s pathway, impairing lymphatic drainage and contributing to osteonecrosis¹¹. However, a systematic investigation of ER stress-related cell death in osteoporosis is still lacking, while the underlying molecular mechanisms and potential therapeutic targets remain to be fully elucidated.

Single-cell RNA sequencing (scRNA-seq) enables high-resolution identification of cellular heterogeneity, allowing for the discrimination of distinct cell subtypes within complex populations. Integrating bulk transcriptomics with scRNA-seq forms a complementary strategy: bulk data reveal overall gene expression trends at the population level, while single-cell data resolve heterogeneity and uncover functional differences among cell subpopulations^12,13. Therefore, this study leverages an integrative approach, combining bulk and single-cell transcriptomic analyses with machine learning and molecular docking, to unveil the core molecular mechanisms and distinct cellular subpopulations associated with endoplasmic reticulum stress-mediated cell death in osteoporosis. Given that the pathogenesis and progression of osteoporosis (OP) are closely associated with osteoclast-mediated bone resorption (PMID: 41530784), and that circulating monocytes serve as the primary precursor cells for osteoclasts (PMID: 41345114) whose transcriptional profiles can reflect the systemic state related to bone resorption, this study adopts an integrated analytical approach. Simultaneously, monocyte/macrophage populations in the bone marrow directly participate in the regulation of bone homeostasis through intercellular interactions (PMID: 41115147), suggesting their potential key role in the pathological metabolism and inflammatory microenvironment of OP. Therefore, this study employs an integrated methodology, combining bulk transcriptomic data from peripheral blood monocytes with single-cell transcriptomic data from bone marrow monocytes, alongside machine learning and molecular docking, to elucidate the core molecular mechanisms and specific cellular subpopulations involved in endoplasmic reticulum stress-mediated cell death in osteoporosis. Our findings offer a conceptual framework for the development of innovative therapeutic interventions aimed at modulating endoplasmic reticulum stress in this disease.

Methods

Data collection

OP-related training dataset (GSE56815), validation dataset (GSE56814), external validation dataset (GSE2208), and scRNA-seq dataset (GSE147287) were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). GSE56815 comprises 40 blood monocyte samples from OP patients (with low bone mineral density (BMD)) and 40 from controls (with high BMD), which were profiled using the GPL96 platform. GSE56814 includes 42 blood monocyte samples from OP patients (with low BMD) and 31 from controls (with high BMD), analyzed with the GPL5175 platform. GSE2208 comprises blood monocyte samples from 9 OP patients (low BMD) and 10 from controls (high BMD), which were profiled using the GPL96 platform. GSE147287 contains freshly isolated bone marrow-derived monocyte samples from the femoral head of 1 OP patient (T-score: -3.3 at lumbar spine, -3.7 at the total hip), with osteoarthritis samples excluded. A key limitation of this dataset is its small sample size (n = 1) and the absence of healthy control samples. The profiling was processed using the GPL24676 platform. A total of 2903 ERS-related genes were obtained by searching the GeneCards database (https://www.genecards.org/) for ER stress (Additional file 1). A total of 1548 programmed cell death (PCD)-related genes were downloaded from the literature¹⁴ (Additional file 2). The access time for all the above data was August 1, 2025.

To address potential confounding factors, we supplemented the key clinical baseline characteristics for GSE56815, GSE56814 and GSE2208. The detailed information, including gender distribution, menopausal status, and BMD, is provided in Additional file 3–5. Furthermore, we performed a baseline characteristic balance analysis between the OP and control groups within each dataset to ensure comparability (Additional file 6–8).

The GSE56815, GSE56814 and GSE2208 dataset was downloaded from the GEO database using the getGEO() function from the GEOquery package (v2.73.4), and the expression matrix was extracted using the exprs() function from the Biobase package (v2.62.0). After probe mapping and data cleaning (retaining the probe with the highest expression for each gene and removing missing values), quantile statistics confirmed that the data had already undergone RMA normalization and log2 transformation; therefore, no further normalization was performed. Finally, based on the phenotypic information, samples were assigned to groups (Control/Disease) and protein-coding genes were filtered, yielding the standardized expression matrix for subsequent analysis.

Analysis of differentially expressed genes (DEGs)

Differential expression analysis between the OP and control groups was performed on the training dataset gene expression matrix using the limma package (v3.58.1)¹⁵, with an adjusted p-value < 0.05 set as the significance threshold¹⁶. The resulting DEGs were visualized in a volcano plot generated with ggplot2 (v3.5.1)¹⁷. Additionally, the top 10 up- and down-regulated DEGs, ranked by absolute log2 fold-change, were displayed in an expression heatmap created using the ComplexHeatmap package (v2.14.0)¹⁸. Candidate genes were identified by intersecting the DEGs with genes associated with endoplasmic reticulum (ER) stress and programmed cell death (PCD), which was conducted and visualized via the VennDiagram package (v1.7.3)¹⁹.

Protein–protein interaction (PPI) network construction and enrichment analysis

A protein–protein interaction (PPI) network for the candidate genes was constructed using the STRING database (confidence score ≥ 0.4) to elucidate the functional interactions among the encoded proteins. The resulting network was visualized with the circlize package (v0.4.16)²⁰. Furthermore, to interpret the biological roles of these genes, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway^21–23 enrichment analyses were performed using the clusterProfiler package (v4.8.3)²⁴, with terms and pathways considered significant at p < 0.05.

Machine learning

To refine the candidate genes and identify those with pivotal roles in OP, we employed three distinct machine learning algorithms on the training dataset. The Least Absolute Shrinkage and Selection Operator (LASSO) regression was implemented using the glmnet package (v4.1-8)²⁵ with fivefold cross-validation, and genes retained at the optimal penalty parameter (lambda.min) were designated as LASSO genes. The Support Vector Machine-Recursive Feature Elimination (SVM-RFE) algorithm, executed via the caret package (v6.0-94)²⁶ with fivefold cross-validation, selected the feature set yielding the highest predictive accuracy as SVM-RFE genes. A Random Forest (RF) model was built with the randomForest package (v4.7-1.1)²⁷, and the top 10 genes, as ranked by the "minimum error regression trees" criterion, were defined as RF genes. Finally, the overlapping genes from these three sets were identified as key intersection genes using the VennDiagram package (v1.7.3).

Gene expression level analysis

Differential expression analysis of the intersection genes between the OP and control groups was performed using the Wilcoxon rank-sum test. Genes demonstrating a consistent direction of expression change and a statistically significant difference (p < 0.05) in both the training and validation datasets were subsequently defined as final biomarkers.

Construction and evaluation of a nomogram

To assess the collective predictive utility of the identified biomarkers for OP, a nomogram was developed using the rms package (v6.8.1)²⁸ based on the training dataset. In this model, each biomarker is assigned a points score proportional to its expression level. The summation of these individual scores yields a total points value, from which the probability of OP incidence can be directly read; a higher total score corresponds to a greater predicted risk. The model’s calibration was evaluated by plotting a calibration curve (using the rms package), where a slope closer to 1 indicates superior agreement between predicted and observed outcomes. The discriminatory power of the nomogram was quantified by Receiver Operating Characteristic (ROC) analysis with the pROC package (v1.18.5)²⁹, reporting the Area Under the Curve (AUC). An AUC value > 0.7 and ≠ 1 was considered indicative of satisfactory model performance, with higher values denoting better prediction. Furthermore, the clinical applicability of the nomogram was appraised using Decision Curve Analysis (DCA) implemented via the rmda package (v1.6)³⁰, which estimates the net benefit across a range of decision thresholds.

Gene set enrichment analysis (GSEA)

To delineate the signaling pathways associated with the biomarkers, we first profiled their co-expression networks. Specifically, Spearman correlation analyses between each biomarker and all other genes were performed across all training set samples using the psych package (v2.4.3)³¹. Genes were then ranked in descending order based on the derived correlation coefficients for each biomarker. For the Gene Set Enrichment Analysis (GSEA), the C2: KEGG subset from the Molecular Signatures Database (MSigDB) was retrieved as the reference gene set using the msigdbr package (v7.5.1)³². Enrichment pathways for each biomarker were subsequently characterized using GSEA implemented in the clusterProfiler package (v4.8.3). A signaling pathway was considered significantly enriched with a p-value < 0.05 and an absolute Normalized Enrichment Score (|NES|) > 1.

Localization analysis

The chromosomal locations of the identified biomarkers were mapped using the RCircos package (v1.2.2)³³.

Molecular regulatory network construction

To elucidate the upstream regulatory mechanisms of the biomarkers, putative microRNAs (miRNAs) targeting these genes were predicted using the microcosm database accessed via the multiMiR package (v1.16.0)³⁴. Concurrently, potential transcription factors (TFs) were identified using the mirNet database. A comprehensive miRNA-TF-biomarker regulatory network was subsequently visualized using Cytoscape software (v3.10.2)³⁵.

Analysis of immune cell infiltration

In OP, alterations in immune status were induced, which gave rise to a chronic low-grade inflammatory phenotype. Immune cells were shown to interact with bone cells through direct contact or paracrine mechanisms, with various cytokines and other mediators released. These mediators affected the balance between bone formation and resorption, thereby exacerbating bone destruction and contributing to the pathogenesis of the disease³⁶. The immune cell infiltration landscape was profiled in the training dataset using the single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm implemented in the GSVA package (v1.50.0)³⁷, which estimated the relative abundances of 28 immune cell types. The gene signatures for these 28 immune cell types were obtained from the literature ³⁸ (PMID: 28052254). Before ssGSEA, the data had been standardized. During ssGSEA, the built—in normalization in the gsva() function was applied. The principle of this normalization is to standardize the ssGSEA scores of each sample according to the mean and standard deviation of all gene set scores in that sample, eliminating baseline differences between samples and making the scores of different gene sets comparable. The Wilcoxon rank-sum test was then applied to identify immune cell populations that were significantly dysregulated (p < 0.05) between the OP and control groups. Then, the Benjamini-Hochberg (BH) method was applied for multiple testing correction of the Wilcoxon p-values to control the false discovery rate (FDR). Furthermore, Spearman correlation analysis, performed with the psych package (v2.4.3), was used to investigate the interrelationships among these differential immune cells and their associations with the biomarkers, applying thresholds of |correlation coefficient (cor)|> 0.3 and p < 0.05.

Drug prediction and molecular docking

To pinpoint candidate therapeutics capable of targeting the identified biomarkers, we interrogated the Drug-Signature Database (DsigDB) and visualized the resultant drug-biomarker interactions using Cytoscape (v3.10.2). The binding affinities between the top candidate drugs and their corresponding biomarker-encoded proteins were subsequently evaluated via molecular docking using the CB-DOCK2 web server. The three-dimensional structures of the compounds were retrieved from PubChem, while protein structures for the biomarkers were obtained from the Protein Data Bank (PDB), prioritizing entries with the highest resolution. For biomarkers lacking a solved structure, homology models were generated using AlphaFold. Molecular docking was then performed with CB-DOCK2, and the resulting poses were visualized. Consistent with conventional criteria, a docking score of less than -5 kcal/mol was considered indicative of strong binding potential.

The scRNA-seq analysis

To delineate the single-cell expression patterns of the biomarkers and reconstruct the developmental trajectories of key cell types in osteoporosis, we performed a comprehensive analysis of the scRNA-seq dataset using Seurat (v5.1.0). Initial quality control involved creating a Seurat object (min.cells = 3, min.features = 200) and filtering doublets with scDblFinder (v1.16.0). We retained cells expressing between 200 and 5000 genes, excluded genes detected in fewer than 200 or more than 20,000 cells, and removed cells with mitochondrial gene content exceeding 20%. The top 2000 highly variable genes were then identified using the FindVariableFeatures function (selection.method = “vst”). After dataset integration with IntegrateData, we performed principal component analysis (PCA) on these variable genes. Significant principal components (p < 0.05) were selected for downstream clustering. Cells were clustered using the FindNeighbors and FindClusters functions (resolution = 0.3) and visualized via UAP. Cluster-specific marker genes were identified with FindAllMarkers (logfc.threshold = 0.25, min.pct = 0.1, only.pos = TRUE). Cell types were annotated by referencing both the SingleR package (v2.0.0) and the CellMarker 2.0 database, guided by established literature. Subsequently, to evaluate the stability of the cell clustering results, sensitivity analyses were performed with resolution = 0.2 and resolution = 0.5.

Screening of key cells

To screen for key cells, based on the scRNA-seq dataset, the distribution of biomarkers in the annotated cells and their expression levels across each annotated cell type were first presented. Cells with high expression levels of the biomarkers were selected as key cells. Subsequently, functional enrichment analysis of all cells was performed using ReactomeGSA (v 1.02.0)³⁹ to explore their biological pathways (adjusted p < 0.05). The top 10 pathways with the most significant adjusted p-values were displayed.

Cells communication

To systematically map intercellular signaling, we performed cell–cell communication analysis using the annotated scRNA-seq dataset. The CellChat package (v1.6.1)⁴⁰ was employed to computationally infer interactions between defined cell populations based on the expression profiles of receptor-ligand pairs. Significant interactions were identified using the following thresholds: p < 0.05 and a combined expression level (log₂-mean [Molecule 1, Molecule 2]) ≥ 0.1.

Pseudotime analysis

To understand the developmental trajectories of key cells and the expression changes of biomarkers in key cells, dimensionality reduction and clustering were first performed on key cells based on the scRNA-seq dataset (as in previous steps, details omitted here). With the resolution set to 0.3, a UMAP clustering plot was generated. Following this, cluster-specific marker genes that were highly expressed in key cell clusters were identified using the FindAllMarkers function in Seurat (v5.1.0), with the following thresholds applied: min.pct = 0.25 and only.pos = TRUE. Pseudotime analysis was then performed on the key cell subsets using Monocle 2 (v 2.28.0)⁴¹. All cells in the cell subsets were ordered according to their pseudotime and projected onto one root and two branches, enabling pseudotime trajectory analysis. Additionally, the expression changes of biomarkers in the cell subsets were visualized.

Correlation analysis

To explore the potential functional associations of the identified biomarkers with key biological processes, established marker genes for key cell type differentiation, ERS/unfolded protein response (UPR), and apoptosis/PCD were compiled from relevant literature to form three independent gene sets^42–49 (PMID: 37948991 30210899 26073941 29592897 30978349 35662627 36246375 28286085). From the apoptosis/PCD gene set, genes were separately subjected to correlation analysis with each biomarker. For each biomarker, genes exhibiting a significant correlation (p-value < 0.05) were identified and ranked by the absolute value of their correlation coefficient. The top 10 genes for each biomarker were selected, and the intersection of these two top-10 lists was taken to obtain a final, consensus set of apoptosis/PCD markers. Finally, correlation analysis was performed among this final set of apoptosis/PCD markers, the differentiation markers, the ERS/UPR markers, and the biomarkers in GSE147287 dataset.

TF’s regulatory analysis

To investigate the regulated TFs among different subpopulations of key cells, Variation of Information-based Pathway Enrichment in RNA (VIPER) analysis was performed. The analysis was conducted using Dorothea (v 1.12.0)⁵⁰, which helps reveal the relationship between gene expression patterns and regulatory networks by calculating the activity of TFs. A heatmap was used to display the core regulatory TFs in each subpopulation of candidate key cells. These TFs were ranked from high to low according to their activity, and the top 10 TFs with the highest activity were selected for display.

Experimental validation

To experimentally validate the differential expression of the identified biomarkers, reverse transcription quantitative polymerase chain reaction (RT-qPCR) was performed. Clinical peripheral blood samples were obtained from the First Affiliated Hospital of Yunnan University of Chinese Medicine (Yunnan Provincial Hospital of Chinese Medicine) with approval from the Institutional Ethics Committee (Approval No. 2025-KY-018-01), and all procedures involved in this study strictly adhered to the ethical principles outlined in the Declaration of Helsinki for research involving human participants. Informed consent has been obtained from the patient. All participants were fully informed of the study’s purpose, procedures, potential risks, and benefits. After thorough communication with the researchers and a complete understanding of the information, they voluntarily signed a written informed consent form. The detailed clinical information and baseline characteristic balance analysis of the samples used in this experiment is provided in Additional file 9–10. Total RNA was isolated using the TRIzol method, and its concentration was quantified with a NanoPhotometer N50. cDNA synthesis and quantitative PCR were subsequently carried out using commercial master mixes (HP All-in-one qRT Master Mix II RT203-Ver.1, Kunming Younggen Biotechnology Co., Ltd.; SwsScript All-in-One First-strand-cDNA-synthesis SuperMix for qPCR, Servare Company). The primer sequences used are provided in Table 1. Relative gene expression levels of the biomarkers were calculated using the 2^-ΔΔCT method.

Table 1.

Table of primer sequence.

Primer	Sequence
CAMKK2 F	GCAGGGTCAGTGAGACATCC
CAMKK2 R	TTGGATCCCCCAGCTGGATA
DAPK3 F	GCACGACATCTTCGAGAACAA
DAPK3 R	CTTAGAGTGCAGGTAGTGAACG
GAPDH F	ATGGGCAGCCGTTAGGAAAG
GAPDH R	AGGAAAAGCATCACCCGGAG

Statistical analysis

All statistical and bioinformatic analyses were conducted using R (v 4.3.1). Data visualization was performed with GraphPad Prism 10. For group comparisons, p-values were derived from t-tests or Wilcoxon rank-sum tests, as appropriate, with a p-value < 0.05 considered statistically significant.

Results

Screening and enrichment analysis of candidate genes

In the initial screening for candidate genes, a total of 391 differentially expressed genes (DEGs) were identified between the OP and control groups (adjusted p < 0.05). This set comprised 269 up-regulated and 122 down-regulated genes in the OP group (Fig. 1a). A density distribution analysis of the log₂ fold-change (log₂FC) values for these DEGs revealed that the vast majority (99.5%) exhibited |log₂FC|≤ 0.5, with only 2 genes (0.5%) showing |log2FC|> 0.5. This distribution supports the stringency of our differential expression analysis (Fig. 1b). Then, the intersection of the 391 DEGs, 2903 ER stress-related genes, and 1548 PCD-related genes was taken, and 28 candidate genes were obtained (Fig. 1c and Additional file 11). Then, a PPI network was constructed for the candidate genes (confidence score ≥ 0.4). It could be seen that genes such as EGFR and CTNNB1 had relatively close interaction relationships with other genes (Fig. 1d). Finally, enrichment analysis of the candidate genes (p < 0.05) identified 950 significantly enriched GO terms. These comprised 817 biological processes, primarily associated with autophagy regulation, apoptotic signaling pathway regulation, and cellular response to oxidative stress; 70 cellular components, notably enriched in vesicle lumen, secretory granule lumen, and transcription repressor complex; and 63 molecular functions, prominently featuring RAGE receptor binding, Toll-like receptor binding, and oxidoreductase activity acting on metal ions (Fig. 1e and Additional file 12). Furthermore, KEGG pathway analysis revealed 64 significantly enriched signaling pathways. These encompassed several key processes, notably cellular senescence, chemical carcinogenesis—reactive oxygen species, and the FoxO signaling pathway (Fig. 1f and Additional file 13). The enrichment of GO biological processes and KEGG pathways—particularly those involving autophagy regulation, apoptotic signaling, cellular response to oxidative stress, cellular senescence, ROS-related chemical carcinogenesis, and the FoxO signaling pathway—suggests a potential role for these candidate genes in modulating bone metabolic balance during osteoporosis development. This regulation likely occurs through influencing endoplasmic reticulum stress and cell death processes (e.g., apoptosis) in bone cells. These findings provide valuable insights into the involvement of ER stress and cell death-related mechanisms in OP progression.

Fig. 1.

Screening of candidate genes and construction of PPI network and enrichment analysis. (a) Volcano plotand heatmap of OP DEGs (adjusted p 0.05; up-regulated: 269 genes, down-regulated: 122 genes); (b) Densitydistribution of the log2FC values for OP DEGs (|log2FC| ≤ 0.5: 389 genes, |log2FC| 0.5: 2 genes); (c) Venn diagram ofthe intersection of DEGs, ES stress-related genes, and PCD-related genes (candidate genes: 28); (d) Construction ofPPI network (confi dence score ≥ 0.4); (e) GO enrichment analysis (p 0.05); (f) KEGG enrichment analysis (p 0.05).

Screening of biomarkers

To refine the candidate genes and identify biomarkers closely associated with OP pathogenesis, we employed a machine learning-based screening strategy. Initially, LASSO regression analysis was applied, yielding 17 feature genes [log(lambda min) = -2.9233] (Fig. 2a and Additional file 14). Through SVM-RFE analysis, 27 SVM-RFE genes were selected (Fig. 2b and Additional file 15). The Random Forest (RF) model was constructed using an optimal tree threshold of 91, which corresponded to the lowest error rate, and the top 10 genes were subsequently selected as the final RF gene set (Fig. 2c and Additional file 16). After taking the intersection of the three sets, 10 intersection genes were obtained, namely FOXO3, BRSK2, UQLN2, RAC1, CAMKK2, PTEN, S100A9, APPL1, DAPK3, and PHLDA3 (Fig. 2d). Subsequently, an analysis of the expression levels of these intersection genes was performed, and it was found that the expression levels of CAMKK2 and DAPK3 were down-regulated in OP in both the training dataset and the validation dataset (Fig. 3a,b). Based on these findings, CAMKK2 and DAPK3 were designated as the definitive biomarkers for all subsequent investigations. To experimentally validate the bioinformatics predictions, we assessed the expression levels of CAMKK2 and DAPK3 using RT-qPCR. The results confirmed that both genes were significantly downregulated in the OP group compared to the controls (p < 0.05; Fig. 3c), thereby providing independent experimental corroboration for our computational findings.

Fig. 2.

Screening of intersection genes. (a) Cross-validation plot and penalty coeffi cient plot for parameterselection in LASSO analysis [log(lambda min) = -2.9233]; (b) Curves of classifi cation results for accuracy and errorrate in SVM-RFE analysis; (c) Curves of model error rate and ranking plot of feature importance in RF analysis; (d)Venn diagram of the intersection of LASSO genes, SVM-RFE genes, and RF genes (intersection genes: 10).

Fig. 3.

Analysis of gene expression levels of biomarkers. (a) Analysis of gene expression levels in the trainingdataset (p 0.05); (b) Analysis of gene expression levels in the validation dataset (p 0.05); (c) Experimental validationof the biomarkers was conducted using RT-qPCR assays (p 0.05).

Predictive accuracy of biomarkers for OP

A nomogram was subsequently developed to quantify the risk of OP onset based on the biomarker profile. In this model, higher expression scores for CAMKK2 and DAPK3 contributed to an increased total points value, which corresponded to a greater predicted probability of disease. For example, a total score of 81.1 points translated to an OP probability of 29.6% (Fig. 4a). Evaluation of the nomogram revealed a calibration curve closely aligned with the ideal reference line, indicating strong agreement between predicted and observed outcomes. The model demonstrated robust discriminatory power, with an area under the ROC curve (AUC) of 0.786. Furthermore, decision curve analysis (DCA) confirmed the clinical utility of the nomogram, showing a superior net benefit across a wide range of risk thresholds compared to the individual biomarkers (Figs. 4b–d). The independent dataset GSE2208 was used for external validation of the nomogram (Fig. 4e). The nomogram evaluation, ROC curve (AUC = 0.756), and DCA analysis demonstrated that the nomogram model possessed a certain level of accuracy (Fig. 4f–h). In summary, the nomogram exhibited excellent diagnostic and predictive performance, validating its reliability and potential as a practical quantitative tool for the early screening of osteoporosis.

Fig. 4.

Construction and evaluation of the nomogram model. (a) Construction of the nomogram model inGSE56815 dataset; (b) Calibration curve of the nomogram model in GSE56815 dataset; (c) ROC curve of thenomogram model in GSE56815 dataset (AUC = 0.786); (d) DCA analysis of the nomogram model in GSE56815dataset; (e) Valuation of the nomogram model in GSE2208 dataset; (f) Calibration curve of the nomogram model inGSE2208 dataset; (g) ROC curve of the nomogram model in GSE2208 dataset (AUC = 0.756); (h) DCA analysis of thenomogram model in GSE2208 dataset.

Enrichment analysis and construction of regulatory networks of biomarkers

Subsequent Gene Set Enrichment Analysis (GSEA) of CAMKK2 and DAPK3 revealed their significant co-enrichment in multiple signaling pathways (p < 0.05, |NES|> 1). These included cytokine-cytokine receptor interaction, long-term depression, and the vascular endothelial growth factor signaling pathway (Fig. 5a and Additional files 17–18), implicating their potential roles in these biological processes. This suggested that they might be involved in the pathological process of OP through the regulation of these pathways, and in particular, could further link ER stress status and cell death processes by influencing cytokine-mediated inflammatory responses, synapse-related signal regulation, and angiogenesis-related mechanisms, thereby affecting bone metabolic balance.

Fig. 5.

Correlation analysis of biomarkers. (a) GSEA analysis of CAMKK2 and DAPK3 (p 0.05, |NES| 1); (b)Chromosomal localization of CAMKK2 and DAPK3; (c) miRNA-biomarker-TF network.

Chromosomal localization analysis revealed that CAMKK2 was located on chromosome 12, and DAPK3 was located on chromosome 19 (Fig. 5b). For the construction of regulatory networks for the biomarkers, it was found that DAPK3 was regulated by multiple miRNAs (such as hsa-miR-149-5p), whereas fewer miRNAs (such as hsa-miR-345-5p) regulated CAMKK2. TFs such as MLLT1, ZNF76, and HIC1 could regulate both CAMKK2 and DAPK3 simultaneously (Fig. 5c and Additional files 19–20).

Analysis of immune cells in OP

We subsequently investigated alterations in the immune landscape during OP pathogenesis. Analysis revealed that Myeloid-Derived Suppressor Cells (MDSCs) exhibited the highest infiltration level among all cell types in both the OP and control groups (Fig. 6a). Comparative analysis identified three immune cell types with significantly altered abundance: activated dendritic cells, mast cells, and plasmacytoid dendritic cells were all substantially decreased in the OP group compared to controls (p < 0.05; Fig. 6b). Furthermore, correlation analysis demonstrated a significant positive relationship between activated dendritic cells and mast cells (cor = 0.43, p < 0.0001). Both CAMKK2 and DAPK3 showed modest but statistically significant positive correlations with activated dendritic cells (cor = 0.24, p < 0.05; Fig. 6c and Additional file 21). Collectively, these results suggest that immune dysregulation, characterized by specific cellular deficiencies and their relationship with these biomarkers, may contribute to OP pathology, offering new insights into immune regulation in OP development.

Fig. 6.

Analysis of immune infi ltration levels. (a) Enrichment analysis of 28 immune cell types; (b) Analysis ofdiff erential immune cells (diff erential immune cells: 3) (p 0.05); (c) Correlation analysis between biomarkers anddiff erential immune cells (p 0.05).

Molecular docking between CAMKK2, DAPK3, and drugs

To discover candidate drugs for OP treatment, we performed a systematic drug prediction. The results identified multiple compounds, such as captopril and cephaeline, that showed potential for interacting with CAMKK2 and DAPK3 (Additional file 22a and Additional file 23). Subsequently, from a biological perspective, calcitriol, a drug that interacts with CAMKK2 and can promote calcium absorption and bone mineralization, was selected for molecular docking. For DAPK3, danazol was chosen for molecular docking; this is a synthetic androgen derivative that has been used in some studies on postmenopausal osteoporosis to reduce bone loss (through increasing estrogen levels or decreasing bone resorption). The results of molecular docking showed that the binding ability between the drugs and the biomarkers was relatively stable, among which the binding energy of danazol and DAPK3 was -9.1 kcal/mol, indicating that their binding state was good (Additional file 22b-c and Additional file 24). These results suggest potential interactions between CAMKK2 and calcitriol, as well as between DAPK3 and danazol, providing a novel perspective on targeted therapy for OP to some extent. However, given the predictive nature of drug prediction and molecular docking analysis, further experimental validation is required.

Identification of bone marrow-derived mesenchymal stem cells (BM-MSCs) as a candidate cellular player

To complement the bulk transcriptomic findings and investigate the cellular mechanisms underlying OP, we performed single-cell RNA sequencing (scRNA-seq) analysis. After initial quality control of the dataset containing 9654 cells and 20,435 genes, we retained 7497 high-quality cells and all genes for downstream processing (Fig. 7a). We then identified 2000 highly variable genes for dimensional reduction. Based on principal component analysis (PCA) results, the top 20 significant principal components (p < 0.05) were selected for subsequent analyses (Fig. 7b,c). UMAP clustering analysis was then performed, and 14 cell clusters were identified (Fig. 7d). Cell cluster annotation was performed using marker genes, resulting in the annotation of 6 cell types. These included B cells, neutrophils,BM-MSCs, monocytes, T cells, and nucleated red blood cells (NRBCs) (Fig. 7e). A bubble plot was employed to illustrate the distinct specificity of these marker genes, thus confirming the accuracy of the cell type annotations (Fig. 7f). The sensitivity analysis demonstrated that the cell type composition remained relatively stable when using resolution = 0.2 and resolution = 0.5, further supporting the reliability of the annotations (Fig. 7g, Additional file 25). Cell proportion analysis revealed that neutrophils, BM-MSCs, and monocytes accounted for the highest proportion (Fig. 7h). Analysis of biomarker expression revealed that CAMKK2 and DAPK3 were relatively highly expressed in BM-MSCs, so BM-MSCs were selected as candidate cells for subsequent analyses (Fig. 7i,j). Enrichment analysis across annotated cell populations revealed pronounced activity of the TWIK-related acid-sensitive K + channel (TASK) in most cell types, including BM-MSCs. In contrast, although MGMT-mediated DNA damage reversal was highly active in the majority of cells, its activity was distinctly suppressed in BM-MSCs (adjusted p < 0.05; Fig. 7k and Additional file 26). These findings indicate that differential activity in these pathways may contribute to OP pathogenesis by disrupting core BM-MSC functions—including proliferation, differentiation, and DNA damage repair. This perspective offers valuable mechanistic insights for further investigating how cell-type-specific pathway dysregulation influences OP initiation and progression.

Fig. 7.

Cell annotation of scRNA-seq. (a) Data quality control before and after (number of genes, total number ofmRNA molecules, proportion of mitochondrial genes); (b) Screening of highly variable genes; (c) Cell distribution inPCA analysis, Jackstraw plot, and elbow plot; (d) UMAP cell clustering plot; (e) Clustering plot of cell annotation; (f)Bubble plot of marker genes for cell annotation; (g) Line chart of the sensitivity analysis; (h) Enrichment proportionplot of annotated cells; (i-j) Distribution and expression of CAMKK2 and DAPK3 in annotated cells; (k) Enrichmentanalysis of annotated cells.

Specific communications of cell types

Intercellular communication analysis revealed an extensive interaction network among annotated cell types. BM-MSCs demonstrated particularly prominent connectivity, engaging in numerous interactions with heightened communication strength to various immune populations, including monocytes, T cells, and B cells (Fig. 8a). Analysis of receptor-ligand interactions revealed distinct communication patterns. Within the BM-MSC population, interactions were primarily mediated by the ANGPTL4–CDH11 pair. In contrast, ligand-receptor complexes such as MIF–(CD74 + CXCR4) and MIF–(CD74 + CD44) served as potential key mediators across multiple intercellular communication axes, including monocyte-to-NRBC signaling (Fig. 8b). These results indicate that dysregulated intercellular communication—manifested as imbalanced signaling between BM-MSCs and immune cells, along with disrupted ligand-receptor mediated interactions—may contribute to OP pathogenesis by impairing BM-MSC function and disturbing immune microenvironment homeostasis. This insight offers a valuable direction for further elucidating how cell–cell communication networks regulate the initiation and progression of OP.

Fig. 8.

Analysis of intercellular communication. (a) Network diagram of the number and intensity ofcommunication among annotated cells; (b) Bubble plot of the communication probability of receptor-ligand pairs incell communication.

BM-MSCs differentiation trajectory

Subsequently, dimensionality reduction clustering was performed on BM-MSCs, resulting in 6 cell clusters (Fig. 9a). Pseudotime analysis of BM-MSCs revealed that subcluster 0 was primarily localized at the initiation stage of developmental differentiation, subclusters 3 and 4 were in the middle stage of differentiation, subclusters 1 and 2 were at the terminal stage of differentiation, and subcluster 5 was distributed throughout the entire differentiation process. At the state level, states 1, 3, 6, and 7 belonged to the initial state of differentiation, state 5 was in the middle stage of differentiation, and state 4 was in the late stage of differentiation (Fig. 9b). Further analysis of biomarker expression patterns revealed that CAMKK2 exhibited multiphasic expression dynamics during BM-MSC differentiation: it was initially low in the early stage, subsequently increased, then declined, and finally re-increased during the late stage (Fig. 9d). In contrast, DAPK3 expression peaked during early differentiation, declined at the intermediate stage, and subsequently rebounded in the late phase (Fig. 9c). Finally, an activity analysis of TFs in different subclusters of BM-MSCs revealed that TFs such as ZNF277, GTF3A, and THAP4 exhibited strong activity in BM-MSCs subclusters 0 and 3, while their activity was suppressed in subcluster 2 (Fig. 9d). To further explore the potential mechanistic roles of CAMKK2 and DAPK3, correlation analyses were performed with classical marker genes for osteogenic differentiation, ERS/UPR, and apoptosis/PCD. The results showed that both CAMKK2 and DAPK3 exhibited significant positive correlations with the majority of the examined markers, including key osteogenic differentiation genes (BGLAP, ALPL), ERS/UPR genes (HSPA5, ATF4), and apoptosis/PCD-related genes (TNFRSF21, FAIM, PAWR, TNFRSF1A) (Fig. 9e). Taken together, these results indicated that BM-MSCs exhibited distinct stage division and subcluster specificity during the differentiation process. The expression of CAMKK2 and DAPK3 exhibited dynamic, stage-specific alterations during differentiation, concurrent with significant variations in transcription factor (TF) activity across different subclusters. These findings suggest that these TFs may orchestrate the precise progression of BM-MSC differentiation by temporally controlling the expression of key biomarkers such as CAMKK2 and DAPK3. Dysregulation of this process may contribute to aberrant BM-MSC differentiation, thereby potentially underlying the pathogenesis and progression of osteoporosis (OP). The significant positive correlations between CAMKK2 and DAPK3 with established markers for osteogenesis, ERS, and PCD further substantiate their pivotal roles within this regulatory network. These findings provided multi-level evidence for deciphering the molecular regulatory network underlying abnormal BM-MSCs differentiation in OP.

Fig. 9.

Pseudotime analysis of BM-MSCs. (a) Clustering plot of BM-MSCs subclusters; (b) Diff erentiation trajectoryof BM-MSCs (diff erent subclusters, diff erent states); (c) Expression heatmap of CAMKK2 and DAPK3 along thediff erentiation trajectory. (d) Transcription Factors Potentially Regulating Stage-specifi c Biomarker Expression in BM-MSC Diff erentiation. (e) Correlation analyses between biomarkers and classical marker genes for osteogenicdiff erentiation, ERS/UPR, and apoptosis/PCD (p 0.05).

Discussion

Osteoporosis (OP) represents a highly prevalent skeletal disorder, yet long-term pharmacological interventions face challenges including limited efficacy and adverse effects. Growing evidence indicates that sustained endoplasmic reticulum (ER) stress promotes osteoblast apoptosis, thereby playing a critical role in disrupting bone metabolic homeostasis in OP. Consequently, in-depth investigation of ER stress and its associated cell death signaling pathways offers a promising direction for elucidating disease mechanisms and developing novel therapeutic strategies.

This study successfully identified CAMKK2 and DAPK3 as key biomarkers associated with endoplasmic reticulum stress-related cell death in osteoporosis. A nomogram model constructed based on these biomarkers exhibited considerable diagnostic accuracy. Furthermore, Gene Set Enrichment Analysis (GSEA) highlighted significant involvement of these genes in critical pathways, including cytokine-cytokine receptor interaction, long-term depression, and vascular endothelial growth factor signaling. Analysis of the osteoporotic immune microenvironment indicated a weak correlation between the biomarkers and activated dendritic cells. Molecular docking suggested a strong binding affinity between danazol and DAPK3. At the single-cell level, BM-MSCs were identified as candidate cells exhibiting complex communication networks with other cell types. Furthermore, the biomarker expression levels exhibited dynamic alterations during the course of BM-MSC differentiation. In this study, we innovatively integrate the critical pathway of ER stress with complementary bulk and single-cell transcriptomic analyses. This strategy provides a systematic, multi-level perspective on osteoporotic pathogenesis, from molecular mechanisms to cellular behaviors. Our work sheds new light on the dynamic expression profiles of key biomarkers throughout BM-MSC differentiation, as well as their interplay with the immune microenvironment and cell–cell communication, thereby offering a more comprehensive understanding of the disease.

CAMKK2 is a serine/threonine kinase belonging to the Ca²⁺/calmodulin-dependent protein kinase family. Its activation is initiated by Ca²⁺/calmodulin (CaM) binding, which induces a conformational change that exposes the autophosphorylation site and facilitates full activation through autophosphorylation (PMID: 35271909). This kinase primarily regulates crucial physiological processes such as cellular metabolism and immune modulation by phosphorylating key downstream effectors, including AMPKα, CAMK1, and CAMK4⁴². Previous studies have demonstrated that pharmacological inhibition of CAMKK2 promotes fracture healing, stimulates osteoblast formation, and increases bone mass, while CAMKK2-knockout mice exhibit enhanced bone strength⁴³. Furthermore, CAMKK2 accelerates endochondral ossification by modulating the Indian hedgehog signaling pathway⁴³. In the present study, RT-qPCR validation revealed a significant downregulation of CAMKK2 expression in OP samples. Given its central role in metabolism and immune regulation⁴², the downregulation of CAMKK2 in OP may not only reflect an imbalance in bone metabolism but could also influence disease progression through the specific signaling pathways and immune-regulatory networks in which it is involved.

In this study, immune infiltration analysis revealed a significant positive correlation between CAMKK2 expression and activated dendritic cells. Furthermore, GSEA indicated that CAMKK2 was notably enriched in the cytokine-cytokine receptor interaction pathway, suggesting its potential involvement in immune modulation within the bone microenvironment. As a kinase downstream of calmodulin signaling, CAMKK2 acts as an upstream activator of AMPK, implying that it may influence cellular energy metabolism and regulate inflammatory networks through phosphorylation of AMPKα (PMID: 40683882 41573560). Specifically, we hypothesize that downregulation of CAMKK2 could lead to dysregulation of the cytokine network via the "CAMKK2–AMPK" axis (PMID: 41104684), thereby impairing normal regulation of immune cells such as dendritic cells and exacerbating inflammatory status and abnormal bone resorption in the bone microenvironment (PMID: 39289212). Additionally, molecular docking results suggested stable binding affinity between CAMKK2 and Calcitriol, while clinical studies have shown that Calcitriol effectively improves bone mineral density and clinical symptoms in osteoporosis patients (PMID: 41225545). This suggests that Calcitriol may exert its therapeutic effects, at least in part, by targeting CAMKK2 and indirectly modulating downstream AMPK signaling and cytokine networks, thereby alleviating bone loss and improving clinical outcomes in OP patients. These findings offer novel insights and a reference framework for the development of targeted therapies for osteoporosis.

DAPK3 (also known as ZIPK) is a member of the death-associated protein kinase family and functions as a serine/threonine kinase that plays important roles in apoptosis, autophagy, and immune regulation (PMID: 35961135). RT-qPCR results indicated that its expression was significantly downregulated in OP patients. Furthermore, DAPK3 deficiency has been shown to activate endoplasmic reticulum stress (ERS) and upregulate key proteins such as ATF4 and CHOP, thereby promoting osteoblast apoptosis through the PERK–eIF2α–ATF4–CHOP signaling axis^48–50. This suggests that downregulation of DAPK3 may impair osteoblast survival and function via the ERS-apoptosis pathway, ultimately compromising bone formation.

GSEA further revealed that DAPK3 was significantly enriched in the cytokine-cytokine receptor interaction pathway, and its expression positively correlated with levels of activated dendritic cells. Therefore, we propose that downregulation of DAPK3 in OP may exert dual pathological effects: on the one hand, it could directly exacerbate ERS and apoptosis in osteoblasts by enhancing PERK–eIF2α–ATF4–CHOP pathway activity^48–50; on the other hand, it may disrupt the osteoimmune microenvironment by perturbing cytokine networks and impairing the function of immune cells such as dendritic cells through the STING–IFN-β pathway (PMID: 3376742651), thereby indirectly promoting osteoclast activity and bone resorption, collectively aggravating bone loss (PMID: 41477218).

Additionally, studies have reported that Danazol is associated with increased lumbar bone mineral density, suggesting its therapeutic potential for OP (PMID: 8941057, 7750585). In this study, molecular docking indicated a strong binding affinity (− 9.1 kcal/mol) between Danazol and DAPK3. These findings imply that Danazol may act by targeting DAPK3, potentially inhibiting excessive activation of the PERK–eIF2α–ATF4–CHOP pathway to alleviate ERS and apoptosis in osteoblasts. At the same time, its binding could restore DAPK3-mediated regulation of the STING–IFN-β pathway, improve dendritic cell function, and remodel the osteoimmune microenvironment, ultimately reducing bone resorption and promoting bone formation, thereby exerting therapeutic effects against OP.

BM-MSCs, the primary cellular source of osteoblasts, can undergo stepwise differentiation into functional osteoblasts by receiving signals from the microenvironment and activating downstream transcriptional networks (PMID: 41457238). Existing studies indicate that the imbalance between osteogenic and adipogenic differentiation of BM-MSCs is one of the key factors contributing to inadequate bone formation in osteoporosis. For example, under glucocorticoid exposure, upregulated SENP3 in BM-MSCs can promote adipogenic differentiation while inhibiting osteogenesis by affecting the post-translational modifications (e.g., de-SUMOylation) of key differentiation proteins such as PPARγ2[54]. On the other hand, ROS induced by oxidative stress can not only upregulate SENP3 expression but also directly disrupt endoplasmic reticulum (ER) homeostasis, triggering ER stress (ERS) and impairing the protein-folding environment of the ER, thereby activating the unfolded protein response (UPR)[55]. Persistent signaling from ERS and UPR pathways can further interfere with osteogenic/adipogenic transcriptional programs, leading to differentiation imbalance of BM-MSCs, impaired bone formation, and consequently reduced bone mass and increased fracture risk (PMID: 40682619, 38744219). Our single-cell analysis indicates that BM-MSCs are potentially key cells in the pathogenesis of osteoporosis, and their characteristics under OP conditions are closely associated with sustained endoplasmic reticulum stress (ERS). The underlying molecular mechanisms involve the synergistic dysregulation of multiple pathways. In this study, functional enrichment analysis revealed significantly suppressed activity of the MGMT-mediated DNA damage reversal pathway in BM-MSCs, which may lead to the accumulation of DNA damage and increased genomic instability, thereby exacerbating ER proteotoxicity and triggering a persistent unfolded protein response (PMID: 38837103). This provides an upstream molecular basis for the functional decline of BM-MSCs. Simultaneously, the activity of the TWIK-related acid-sensitive K⁺ channel (TASK) pathway was markedly elevated in BM-MSCs. As key regulators of intracellular ion homeostasis, aberrant TASK channel activity may disrupt the balance of intracellular potassium and calcium ions, interfere with ER calcium storage homeostasis, and consequently induce or aggravate ERS (PMID: 40412718, 25998740). Furthermore, disturbances in ion homeostasis can also affect energy metabolism and mitochondrial function, forming a positive feedback loop of ER-mitochondrial dysfunction[56]. In summary, abnormal activities of the MGMT-mediated DNA damage reversal and TASK pathways may synergistically intensify endoplasmic reticulum stress in BM-MSCs at both the genomic stability and organelle homeostasis levels. Together, they impair the self-renewal, differentiation potential, and adaptive capacity of these cells to microenvironmental stress, thereby weakening bone formation and driving the pathological progression of OP (PMID: 40414869).

The biomarkers CAMKK2 and DAPK3 may serve as key nodes linking the aforementioned ERS pathways to BM-MSC dysfunction. Pseudotime analysis indicated that both exhibit dynamic and stage-specific expression patterns during BM-MSC differentiation, suggesting that they may play regulatory roles at specific checkpoints of the differentiation process. Furthermore, correlation analysis revealed significant positive correlations of CAMKK2 and DAPK3 with genes involved in osteogenic differentiation, ERS/UPR, and programmed cell death (PCD). Combined with their significant downregulation in OP, these findings imply that CAMKK2 and DAPK3 are likely involved in regulating osteogenesis, ERS response, and cell survival in BM-MSCs. Specifically, under OP pathological conditions, factors such as aberrant TASK channel activity and suppression of the MGMT pathway collectively disrupt ion homeostasis and genomic stability in BM-MSCs, thereby inducing and aggravating sustained ERS. This persistent ERS microenvironment may indirectly lead to the downregulation of CAMKK2 and DAPK3 (via circulating factors or other systemic influences). The dysregulated expression of these two biomarkers could subsequently impair the osteogenic differentiation capacity, ERS tolerance, and survival signaling of BM-MSCs. Ultimately, these disturbances may jointly contribute to BM-MSC dysfunction, predisposing the cells to ERS-associated PCD and thereby driving the bone-formation deficit observed in OP (PMID: 40414869, 41201016).

Future studies employing stage-specific knockdown or overexpression experiments will be valuable to elucidate the exact functional contributions of these genes across distinct phases of BM-MSC differentiation. In summary, BM‑MSC dysfunction in osteoporosis originates from the combined effects of multiple stressors in the microenvironment, including oxidative stress, ion channel dysregulation, and impaired genomic repair. Endoplasmic reticulum stress (ERS) serves as a central hub that integrates these upstream pathological signals and, by disrupting the expression and dynamic regulation of key biomarkers such as CAMKK2 and DAPK3, leads to reduced cell viability and increased susceptibility to programmed cell death. Ultimately, this results in an imbalance between adipogenic and osteogenic differentiation. These findings provide a theoretical basis for developing novel OP treatment strategies aimed at improving endoplasmic reticulum function in BM‑MSCs.

Conclusions

This study utilized integrated bioinformatics approaches to explore the pathogenesis of osteoporosis (OP), leading to the identification of CAMKK2 and DAPK3 as novel biomarkers implicated in endoplasmic reticulum stress-associated cell death. Bone marrow mesenchymal stem cells (BM-MSCs) were established as a potential key cellular context, offering fresh perspectives on OP mechanisms and potential treatment avenues. However, it should be noted that this study has several limitations. First, the analyzed sample size is limited, and the single-cell dataset is small and lacks control samples, which may affect the generalizability of the conclusions. Second, no strict fold‑change threshold was applied during the bioinformatic screening, and the clinical samples used for preliminary validation were not fully matched in terms of baseline characteristics. Moreover, subsequent validation at the protein level and functional experiments are lacking. In addition, the publicly available datasets lack detailed clinical indicators, making it difficult to rigorously assess the association between biomarkers and disease phenotypes. The ERS‑ and PCD‑related gene sets used in this study are also relatively broad, which may introduce nonspecific signals. In future studies, multi‑level validation of candidate biomarkers should be performed in larger cohorts with well‑matched controls. Functional mechanisms should be elucidated using genetic manipulation and animal models. Furthermore, integrating more precise pathway annotations and richer clinical phenotypic data will help advance these findings toward clinical translation.

Abbreviations

OP

Osteoporosis

ER

Endoplasmic reticulum

ERS

Endoplasmic reticulum stress

ROS

Reactive oxygen species

scRNA-seq

Single-cell RNA sequencing

GEO

Gene expression omnibus

BMD

Bone mineral density

DEGs

Differentially expressed genes

PPI

Protein–protein interaction

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

LASSO

Least absolute shrinkage and selection operator

SVM-RFE

Support vector machine-recursive feature elimination

RF

Random forest

ROC

Receiver operating characteristic

AUC

Area under the curve

DCA

Decision curve analysis

GSEA

Gene set enrichment analysis

MSigDB

Molecular signatures database

NES

Normalized enrichment score

miRNAs

MicroRNAs

TFs

Transcription factors

ssGSEA

Single-sample gene set enrichment analysis

DsigDB

Drug-signature database

PDB

Protein Data Bank

PCA

Principal component analysis

UMAP

Uniform manifold approximation and projection

BM-MSCs

Bone marrow-derived mesenchymal stem cells

NRBCs

Nucleated red blood cells

TASK

TWIK-related acid-sensitive K+ channel

VIPER

Variation of information-based pathway enrichment in RNA

RT-qPCR

Reverse transcription quantitative polymerase chain reaction

cDNA

Complementary DNA

GIOP

Glucocorticoid-induced osteoporosis

Author contributions

Yifeng Xia: Conceptualization, Data curation, Validation, Visualization, Writing–original draft, Writing–review & editing. Zhongyu Peng: Data curation, Validation, Visualization, Writing–review & editing. Lingrui Zhao: Visualization, Writing–review & editing. Yuan Long: Validation, Writing–review & editing. Renwei Chen: Clinical blood sample collection. Jiahao Dong: Clinical blood sample collection. Meixiang Chu: Conceptualization, Writing–review & editing. Weijie Yu: Conceptualization, Supervision, Writing–review & editing. Tao Chen: Conceptualization, Project administration, Supervision, Writing–review & editing. This work currently described has not been published, is not being considered for publication elsewhere, and its publication was approved by all authors.

Funding

This work was supported by the Joint Fund Project of Yunnan University of Traditional Chinese Medicine and Its Colleges and Departments [Grant Number: XYLH202347]; and the 2025 Key Clinical Specialty in Traditional Chinese Medicine—Orthopedics [Grant Number: Yun Cai She (2025) No. 92]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The data that support the findings of this study are openly available in [Gene Expression Omnibus] at [https://www.ncbi.nlm.nih.gov/geo/], reference number [GSE56815, GSE56814, and GSE147287].

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was approved by the Medical Ethics Committee of Yunnan Provincial Hospital of Traditional Chinese Medicine. The approval number and date of approval are as follows: 2025-KY-018-01 and November 27, 2025. All patients provided written informed consent when clinical samples were collected for RT-qPCR experiments to ensure that the research process was in accordance with ethical norms and that the patients’ rights and xwishes were fully respected. All procedures involved in this study strictly adhered to the ethical principles outlined in the Declaration of Helsinki for research involving human participants. This study strictly adheres to ethical guidelines. The researchers comprehensively and clearly explain the study’s purpose, procedures, risks, and benefits to potential participants to ensure full comprehension. Ample time for consideration and opportunities for questions are provided. Upon obtaining explicit and voluntary agreement, a written informed consent form is jointly signed, with a copy provided to the participant for their records. The entire process respects the participant’s autonomy and safeguards their right to withdraw at any time without giving a reason.