Identification of endothelial INSR as an osteosarcoma-related biomarker and therapeutic target based on weighted gene co-expression network analysis

Abstract

Background

Osteosarcoma (OS) is the most common primary malignant bone tumor characterized by a high rate of metastasis. The clinical efficacy of current targeted drugs for OS is limited, highlighting the urgent need to identify novel therapeutic targets. The study aimed to identify potential drug targets for OS by combining weighted gene co-expression network analysis (WGCNA) with single-cell analysis.

Methods

OS-related genes were obtained from the osteosarcoma database, and prognosis-related genes were identified using The Cancer Genome Atlas (TCGA) data. Gene expression profiles and corresponding clinical data were extracted from TCGA, Genotype-Tissue Expression (GTEx), and Gene Expression Omnibus (GEO) databases. Univariate and multivariate Cox analyses were used to screen the genes associated with OS prognosis. WGCNA was applied to detect genes linked to clinical characteristics in OS patients. Differentially expressed genes (DEGs) were analyzed using the GEO dataset. Single-cell RNA sequencing (scRNA-seq) data were employed to assess the expression pattern of the crucial genes in different types of cells in the tumor microenvironment.

Results

A total of 911 OS-related genes were retrieved from the osteosarcoma database. Prognosis analysis identified 110 prognosis-associated genes (P < 0.05). WGCNA revealed 295 genes related to OS clinical characteristics within the turquoise module. DEG analysis across GEO datasets (GSE12865, GSE16088, and GSE14359) identified 220 DEGs (|log2 (FC)| >1 and P < 0.05). Integration of these analyses identified 16 key genes in OS. Based on the expression profiles of these genes, OS patients were classified into three distinct clusters that exhibited significant differences in survival time, immune scores, and T-cell CD4 memory resting levels (P < 0.05). Insulin receptor (INSR) was identified as an independent risk factor predominantly expressed in endothelial cells.

Conclusion

INSR expression in OS tissues is associated with poor clinical outcomes. Activation or dysregulation of endothelial INSR contributes to OS progression, suggesting that endothelial INSR may represent a novel therapeutic target for OS.

Introduction

Osteosarcoma (OS) is a highly aggressive primary bone malignancy, accounting for 20–34% of all bone tumors [1]. Epidemiologically, OS primarily affects children and adolescents, representing the second leading cause of tumor-related deaths in this age group [2, 3]. Despite advancements in surgical resection, neoadjuvant, and adjuvant chemotherapy, the 5-year survival rate for patients with drug-resistant OS remains largely unchanged [2, 4–6]. Therefore, it is necessary to identify effective biomarkers and novel drug targets for improving the diagnosis and treatment of OS.

Weighted gene co-expression network analysis (WGCNA) is a bioinformatics algorithm that identifies gene modules based on highly correlated expression patterns and associates them with external clinical features. By analyzing module-central genes, WGCNA provides insights into potential therapeutic targets [7]. This approach has been increasingly employed for high-throughput data mining by integrating microarray or RNA sequencing data with clinical information to uncover gene-clinical relationships. For example, a recent study utilized WGCNA to screen metastasis-related gene modules in OS [8]. Another investigation revealed that centrosomal protein 55 (CEP55) functions as a core oncogene in OS progression via activating JAK2/STAT3/MMPs axis, as identified through WGCNA [9]. Additionally, WGCNA has been used to obtain the immune-related gene modules in OS, facilitating prognosis evaluation and immunotherapy response prediction [10]. However, further investigation is required to uncover novel and effective therapeutic targets due to the complex molecular mechanisms underlying OS and the intricate interplay among various cell types within the tumor microenvironment (TME).

This study integrated data from The Cancer Genome Atlas Program (TCGA), the Genotype-Tissue Expression Project (GTEx), and the Gene Expression Omnibus (GEO) with WGCNA to identify genes associated with OS prognosis. The relationship between core gene expression and immune infiltration was further explored using single-cell analysis. The results indicated that the insulin receptor (INSR) is a potential prognostic biomarker and therapeutic target for OS.

Materials and methods

Data source

A total of 911 OS-related genes were obtained from the Osteosarcoma Database (http://osteosarcoma-db.uni-muenster.de/index.php). RNA sequencing data and corresponding clinical information for OS patients (TARGET-OS cohort) and control subjects were obtained from TCGA (https://portal.gdc.cancer.gov/) and the Genotype-Tissue Expression (GTEx) (https://www.gtexportal.org). Univariate and multivariate Cox regression analyses and WGCNA were conducted on the TCGA and GTEx data to identify prognostic and OS-related genes.

Three OS-related gene expression datasets, including GSE12865 (platform: GPL6244), GSE16088 (platform: GPL6244), and GSE14359 (platform: GPL96), and two single-cell RNA sequencing datasets, GSE162454 and GSE168664 (platform: GPL24676), were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). These datasets were analyzed to identify differentially expressed genes (DEGs) between OS and non-cancer samples. The screening criteria for DEGs were |log2 (FC)| >1, P < 0.05.

Weighted gene co-expression network analysis (WGCNA)

The “WGCNA” package in R [7] was used to construct a gene co-expression network and identify hub genes associated with key clinical features, including age and metastasis, by integrating data from the TCGA and GTEx databases (a total of 484 samples). To enhance the specificity and robustness of bioinformatics analysis, the previously identified 911 OS-related genes were used for co-expression network construction. Outlier samples were removed before analysis. A soft-thresholding power of 5 was selected to ensure scale-free topology. A weighted adjacency matrix was constructed, followed by hierarchical clustering of the matrix to identify co-expression modules. The central module was determined based on its connectivity and relevance. Pearson correlation coefficients were calculated to assess the association between each module and clinical features. Genes from the module with the strongest correlation to these features were selected for further analysis. Given that age at onset and distant metastasis of OS are considered independent risk factors for poor OS prognosis [11, 12], survival time, age at diagnosis, and metastasis status were key clinical variables in the analysis.

Consensus clustering analysis

Consensus clustering was performed using the “ConsensusClusterPlus” package in R to classify OS patients based on the expression profiles of 16 key OS-related genes [13]. Clustering was conducted using the “kmdist” algorithm with Pearson distance as the similarity metric. The optimal number of clusters was determined to be k = 3. Survival differences among the resulting patient clusters were evaluated using the “survminer” and “survival”.

Single-cell analysis

Single-cell RNA sequencing datasets GSE162454 and GSE168664 (platform: GPL24676) were obtained from the GEO database. Raw gene expression matrices were processed using the Seurat v3.0 package in R with default parameters. Low-quality genes and cells were filtered out, retaining genes expressed in at least 200 cells and high-quality cells with 500 to 6,000 detected genes. Cells with mitochondrial gene content exceeding 10% were excluded to ensure data integrity. Dimensionality reduction was performed using principal component analysis (PCA), and the top 15 principal components (PCs) were selected for subsequent analysis. Cell clustering was carried out using the “FindNeighbors” and “FindClusters” functions, with the resolution set to 30 dimensions (dim = 30). Cell clusters were visualized using a t-distributed stochastic neighbor embedding (tSNE) scatter plot. Annotation of cell clusters was based on canonical marker genes, allowing for the identification of distinct cell populations. Gene expression patterns were then evaluated across different cell types to explore the cellular context and functional roles of key genes.

Immunoinfiltration analysis

Immunoinfiltration analysis was performed based on the TARGET-OS cohort from the TCGA database, following previously described methodologies [14, 15]. The levels of immune cell infiltration across different sample groups were analyzed using the “ESTIMATE,” “CIBERSORT,” and “Estimate the Proportion of Immune and Cancer cells (EPIC)” algorithms implemented in the IOBR package in R. The “ESTIMATE” package in R was used to calculate the Stromal Score, Immune Score, and ESTIMATE Score, which were then analyzed for their correlations with gene expression across groups. The “CIBERSORT” and “EPIC” packages in R were utilized to examine the associations between gene expression and specific immune cell populations in different groups.

Statistical analysis

All data were expressed as “mean ± standard deviation (SD)”. Student’s t-test was employed to compare two groups, while one-way ANOVA was used for comparisons among more than two groups. Pearson correlation coefficients were calculated to assess correlations. P < 0.05 was considered statistically significant.

Results

Screening of genes associated with OS prognosis

A total of 911 OS-related genes were retrieved from the Osteosarcoma database (http://osteosarcoma-db.uni-muenster.de/index.php). Data from the TARGET-OS cohort in the TCGA database(https://portal.gdc.cancer.gov/) was subsequently analyzed to identify genes associated with OS prognosis. Univariate Cox regression analysis revealed that 110 of the 911 OS-related genes were significantly correlated with patient prognosis (P < 0.05) (Fig. 1).

Fig. 1.

Univariate COX analysis of OS-related prognostic genes (P < 0.01)

WGCNA analysis

WGCNA was conducted using the 911 OS-related genes across integrated datasets from the TCGA and GTEx databases, comprising 484 samples. A soft-threshold power of 5 was selected to ensure a scale-free network topology (Fig. 2A). Five co-expression modules were identified (Fig. 2B), with 295 genes clustered in the turquoise module. This module demonstrated significant correlations with overall survival time (R² = 0.95, P = 2e-215), patient age (10–19 years; R² = 0.78, P = 6e-92), and tumor metastasis (R² = 0.45, P = 7e-24) (Fig. 2C). In the turquoise module, a strong positive correlation was observed between gene significance (GS) and module membership (MM) for age (Cor = 0.94, P < 1e-200) and metastasis (Cor = 0.81, P = 2e-154) (Fig. 2D).

Fig. 2.

WGCNA analysis of OS-related genes associated with age and metastasis. A Scale-free topology fit index and mean connectivity plot for various soft-thresholding powers (softpower = 5). B Clustering of module feature genes. C Correlation analysis between modules and clinical features of OS patients by Pearson correlation coefficients. D Scatter plot shows that the turquoise module had the strongest correlation with 10-19-year-old OS patients (10–19 YO) and metastasis, analyzed using Pearson correlation coefficients

GEO data set filters OS DEGs

DEGs between OS and non-cancer samples were screened using three GEO datasets. GSE12865 identified 4,243 DEGs (pediatric OS samples vs. normal osteoblasts, |log2 (FC)| >1 and P < 0.05) (Fig. 3A), GSE16088 identified 7,657 DEGs (OS samples vs. normal human tissues including liver, lymph node, kidney) (Fig. 3B, |log2 (FC)| >1 and P < 0.05), and GSE14359 identified 1,352 DEGs (OS samples vs. normal osteoblasts, |log2 (FC)| >1 and P < 0.05) (Fig. 3C). A total of 220 overlapping DEGs were identified across all three datasets(Fig. 3D).

Fig. 3.

Identification of DEGs from GEO datasets. A–C Volcano plots of DEGs were identified from GSE12865, GSE16088, and GSE14359 datasets, respectively. (|log2 (FC)| >1 and P < 0.05).

Consensus clustering analysis

Sixteen key genes for OS were identified by intersecting DEGs from the GEO datasets, OS-related genes, and the turquoise module genes from WGCNA (Fig. 4A). These genes included: CTSD, TUBB, S100A4, TNFRSF1A, HSP90B1, STOML2, ANXA5, MCAM, ITGB5, LIMK1, HYAL2, ACTN1, NCAM1, SERPINH1, INSR, and HEY1. Consensus clustering analysis based on the expression profiles of these 16 genes was conducted using OS samples from the TCGA database. Patients were grouped into three distinct clusters: C1, C2, and C3 (Fig. 4B). Prognosis analysis revealed that patients in cluster C1 had the most favorable prognosis, whereas those in cluster C3 exhibited the poorest outcomes (Fig. 4C). The ESTIMATE algorithm was applied to calculate the Stromal, Immune, and ESTIMATE scores across the three clusters. Significant differences in Immune Scores were observed among the clusters (Fig. 4D, P < 0.05). Additionally, immunoinfiltration analysis using the CIBERSORT algorithm demonstrated significant differences in the levels of resting memory CD4 + T cells across clusters C1, C2, and C3 (Fig. 4E, P < 0.05).

Fig. 4.

Consensus clustering analysis of key genes in OS. A Venn diagram shows the crucial gene sets at the intersection of the WGCNA gene set, DEGs in OS, and OS-related genes. B Consensus clustering solution map. C Kaplan - Meier analysis of overall survival of patients. D Stromal Score, Immune Score and ESTIMATE Score of patients with OS in three clusters were analyzed in the box diagram and assessed by one-way ANOVA with OS in three clusters. E Box plot analysis of immune cell infiltration in three clusters of OS patients, assessed by one-way ANOVA. * P < 0.05

Identification of OS-related core genes and analysis of immune cell infiltration

To identify core genes associated with OS prognosis, univariate and multivariate Cox regression analyses were performed on the 16 key genes. The results indicated that the INSR gene was an independent risk factor for OS (Fig. 5A). To further investigate the relationship between INSR expression and immune cell infiltration in OS samples, single-cell transcriptomic data from the GSE162454 dataset were analyzed. The findings demonstrated that INSR expression was associated with immune cell populations (Fig. 5B-C), with predominant expression observed in endothelial cells (Fig. 5D-E). Similarly, analysis of the GSE168664 single-cell transcriptome dataset confirmed that INSR expression was enriched in vascular endothelial tissues (Fig. 5F-G). Based on the TARGET-OS dataset, EPIC immune infiltration analysis showed a positive correlation between INSR expression and cancer-associated fibroblasts (CAFs) and endothelial cells (Fig. 6A, P < 0.01). OS patients were stratified into high and low INSR expression groups based on the median expression level. The high-expression group exhibited significantly greater infiltration of endothelial cells compared to the low-expression group (Fig. 6B, P < 0.01). Moreover, a positive correlation was observed between INSR expression and endothelial cell levels (r = 0.31, P = 9.5e-3) (Fig. 6C). These results align with findings from single-cell analyses of GEO datasets.

Fig. 5.

Identification of OS core genes and immune cell infiltration. A Univariate and multivariate Cox analyses of 16 key genes. B UMAP plot shows the two major cell types (immune cells and stromal cells) identified in the GSE162454 dataset. C UMAP plot shows the 8 cell types identified in the GSE162454 dataset [conventional CD4T cell (CD4Tconv), exhausted CD8 + T cell (CD8 + Tex), exhausted CD8 + T cell (conventional CD4T cell (CD4TCONV), exhausted (CD8 + Tex), exhausted CD8 + Tex. endothelial, fibroblasts, malignant, monocyte or macrophage (Mono/Macro), osteoblasts, and plasmocytes]. D UMAP plot displays INSR gene expression in the GSE162454 dataset. E INSR gene expression in different cell types in the GSE162454 dataset. F UMAP plot shows the 5 cell types identified in the GSE168664 dataset (CD8T, endothelial, fibroblasts, monocyte or macrophage (Mono/Macro), and tprolif). G UMAP plot displays INSR gene expression in the GSE168664 dataset

Fig. 6.

Analysis of INSR and immune cell infiltration based on TARGET-OS cohort. A Heat maps show the correlation between INSR expression and 8 immune cell types (B cells, CAFs, CD4 T cells, CD8 T cells, endothelial cells, macrophages, NK cells, and other cells). B The box diagram shows the difference in the levels of 8 kinds of immune cells in the groups with high and low INSR expression, analyzed using Student’s t-test. C Scatter plot shows the correlation between INSR expression and endothelial cells, analyzed using Pearson correlation coefficients. *P < 0.05 and **P < 0.01

Discussion

The high invasiveness and aggressiveness of OS are major contributors to its poor prognosis. OS cells exhibit a strong preference for pulmonary metastasis. Due to the small size of early metastatic lesions and the limited sensitivity of current diagnostic imaging techniques, many patients are diagnosed at stage IV, which is associated with extremely poor outcomes [16, 17]. For cases with localized disease, conventional treatments achieve a cure rate approaching 70% [18]. However, the 5-year survival rate drops to just 20% for patients with metastatic OS [19]. Previous studies have shown that the metastasis status of OS significantly impacts treatment response and plays a crucial role in clinical decision-making [20]. Current targeted therapies for OS primarily focus on the vascular endothelial growth factor (VEGF) and platelet-derived growth factor (PDGF) pathways, along with their respective receptors. These pathways are key regulators of angiogenesis, tumor growth, and metastasis, and their aberrant expression in OS has been associated with an increased risk of lung metastasis and poor survival outcomes [21]. Clinically evaluated targeted agents include cabozantinib, lenvatinib, regorafenib, anlotinib, sorafenib, and imatinib [21–24]. Despite these advances, the emergence of drug resistance remains a major obstacle in OS treatment. Therefore, identifying novel and effective therapeutic targets is urgently required to improve the treatment outcomes of OS.

This study identified 16 key genes related to OS prognosis with abnormal expression characteristics. WGCNA revealed strong correlations between these genes and key clinical traits, including metastasis and age. Cluster analysis further demonstrated significant associations of these genes with prognosis and immune infiltration. Among the 16 key genes, eight were identified as prognostic risk factors and seven as prognostic protective factors, many of which have been previously reported to participate in OS progression. Specifically, S100A4 is overexpressed in various cancers, including OS. Immunohistochemical and clinicopathological analyses have shown that high S100A4 expression is significantly correlated with advanced clinical stage, distant metastasis, and reduced overall and disease-free survival [25]. Heat shock protein 90 (HSP90) has emerged as a promising target for cancer therapy, and genome-wide association studies have suggested that HSP90B1 may serve as a diagnostic biomarker for OS [26]. Stomatin-like 2 (STOML2) has been linked to increased metastatic potential and mortality risk in Ewing’s sarcoma [27]. Similarly, elevated levels of melanoma cell adhesion molecule (MCAM) are strongly associated with poor OS prognosis and are known to regulate tumor cell migration [28]. Nerve cell adhesion molecule 1 (NCAM1, also known as CD56) plays a critical role in the regulation of cancer development [29, 30]. Studies have shown that NCAM1 is associated with prognosis and recurrence in patients with Ewing’s sarcoma, highlighting its potential as a biomarker to refine treatment stratification and enable personalized therapies [31]. Serpin Family H Member 1 (SERPINH1) has emerged as a key biomarker in OS, demonstrating strong predictive value for the efficacy of immunotherapy, chemotherapy response, and clinical outcomes in patients with OS [32]. Hairy/enhancer of split-related with YRPW motif protein 1 (HEY1), a transcription factor regulated by the Notch signaling pathway, has been shown to promote OS cell invasion and pulmonary metastasis [33]. These findings highlight the dysregulation of the 16 identified genes as critical contributors to OS progression. Their dual roles as prognostic markers and modulators of malignant behavior underscore the need for further functional investigations to validate their therapeutic and clinical relevance.

INSR is mainly expressed in the liver, adipose tissue and skeletal muscle and is involved in insulin metabolism. However, INSR is also highly expressed in various human malignancies, including tongue squamous cell carcinoma, colon cancer, breast cancer, and glioma [34–37]. In sarcomas, tumor cell proliferation, survival, and resistance to mammalian target of rapamycin complex 1 (mTORC1) inhibitors have been shown to depend on the insulin-like growth factor 1 receptor (IGF1R) and/or INSR signaling pathways [38]. Additionally, INSR is overexpressed in OS cells and linked to enhanced glycolytic activity [39]. In the present study, INSR emerged as the only independent prognostic risk factor among the 16 key genes, underscoring its pivotal role in OS progression. TME is closely related to tumor immunosuppression, cancer cell metastasis, drug resistance, and response to targeted therapies [40]. The TME comprises tumor cells, infiltrating immune cells, CAFs, extracellular matrix, endothelial cells, and various signaling molecules [40]. Traditional bulk sequencing techniques analyze mixed cell populations, averaging gene expression signals across heterogeneous cell types and consequently obscuring critical cell-specific information [41]. In this study, single-cell RNA sequencing data from OS tissues were analyzed, revealing that INSR expression is predominantly enriched in endothelial cells. INSR expression was also associated with immune cell infiltration and CAFs. Given that OS cells primarily metastasize through hematogenous and lymphatic routes [42], endothelial and lymphatic endothelial cells are considered essential mediators of pathological angiogenesis and lymphangiogenesis, contributing to disease progression [43]. Notably, previous studies have reported that endothelial INSR promotes sprouting angiogenesis via activation of the VEGFA–ERK1/2 signaling pathway [44]. Consistently, based on the analysis of pan-cancer scRNA-seq data, a recent study reports that vascular endothelial growth factor receptor 1 and 2 (VEGFR1/2) are highly expressed in INSR + endothelial cells across multiple human malignancies, which further implies the potential role of INSR as a novel antiangiogenic target [45]. Additionally, increasing studies suggest that the reciprocal action between endothelial cells and OS cells facilitates disease progression. Specifically, endothelial cells secret Von Willebrand factor, which promotes the epithelial-mesenchymal transition of OS cells [46]. Endothelial cells also activates lysophosphatidic acid receptor signaling of OS cells to stimulate aggressiveness [47]. It deserves further explore whether INSR potentiates these molecular actions to drive OS progression in the following studies. The data in the current work also suggested that INSR participated in the abnormal immune microenvironment of OS. Although immunotherapy has achieved remarkable success in several solid tumors, its efficacy in OS remains limited due to low immunogenicity and highly immunosuppressive TME [48]. Even though there are few studies reporting the role of INSR in modulating immune escape of OS cells, multiple studies have implied its critical role in tumor immunity. Insulin signaling in endothelial cells limits leukocyte adhesion via modulating C-X-C motif chemokine receptor 4 [49], probably leading to the immunosuppressive TME. Collectively, these findings suggest that abnormal INSR expression may facilitate OS metastasis through angiogenesis, lymphangiogenesis, and immunosuppression mechanisms. Targeting endothelial INSR may represent a promising therapeutic strategy to inhibit OS metastasis, enhance sensitivity to immunotherapy, and ultimately improve patient outcomes.

Limitations

This study is primarily based on in silico analyses, and the functional role of INSR in regulating the malignant phenotype of osteosarcoma and endothelial cells remains to be experimentally validated. Future studies employing cellular and animal models, and clinical samples, are needed to confirm its biological significance. For example, a co-culture system of vascular endothelial cells and OS cells, with gain-of-function and loss-of-function models of INSR is needed to validate the regulatory effects of INSR in endothelial cells on the metastatic potential of OS cells. The correlation between the expression of INSR in endothelial cells and the infiltration of immune cells can be verified with multiplex immunofluorescence, or scRNA-sequencing based on independent samples. Additionally, the prognostic value of INSR should be verified using clinical specimens from larger, multicenter cohorts to strengthen the translational relevance of these findings.

Conclusion

INSR was identified as a core gene associated with OS progression and shows potential as both a diagnostic biomarker and a therapeutic target. However, the biological implications of its aberrant expression in endothelial cells within OS tissues warrant further investigation.

Author contributions

Conceived and designed the experiments: Xu Yingkai & Wang Lei; Bioinformatics and data analysis: Xu Yingkai and Jiang Jianfeng; Drafted the paper: Xu Yingkai, Hu Zhiyong & Zhou Zhifeng. All authors read and approved the final manuscript.

Funding

This work was financially supported by the Natural Science Foundation and the Social Livelihood Science and Technology Program of Nantong (MSZ2023217).

Data availability

The data used to support the findings of this study are available from the corresponding author upon request.

Declarations

Ethics approval and consent to participate

No humans and animals were involved in the present study.

Competing interests

The authors declare no competing interests.