Comprehensive investigation of hypoxia-associated genes and their predictive significance in osteosarcoma Authors: Shangfang Xie, Wenyao Yu, Runye Lin, Songjian Xin Published: 2026 ABSTRACT Highlights•Our research findings offer valuable insights into the underlying mechanisms of osteosarcoma.•We identified key transcription factors that are associated with patient prognosis and developed a predictive model, which categorized patients based on their risk profiles.•These key regulators offer promising avenues for predicting patient survival and serve as potential therapeutic targets for treatments specifically targeting hypoxia in osteosarcoma.Keywords: Osteosarcoma, Hypoxia, Single-cell analysis, Prognostic model, Cell-cell communication Background Osteosarcoma, a primary malignant bone tumor, stands as the most prevalent of its kind, characterized by its notorious aggressiveness. This devastating disease primarily targets adolescents and individuals over the age of 60, posing a significant health challenge [1]. Surgical intervention combined with radiotherapy is currently considered the preferred treatment for osteosarcoma, but for patients with lung metastases, the five-year survival rate is only approximately 20% [2]. Despite the refinement of surgical methods and the implementation of the MAP chemotherapy regimen, which consists of intensive treatments with methotrexate, doxorubicin, and cisplatin, high-grade osteosarcoma continues to show considerable resistance to chemotherapy. This resistance has resulted in a stagnation of therapeutic outcomes over the past few decades [3]. Osteosarcoma patients have benefited from the progress made in chemotherapy and neoadjuvant chemotherapy treatments. Currently, the commonly used chemotherapeutic drugs in clinical practice include cisplatin, doxorubicin, methotrexate, and ifosfamide [4]. While chemotherapy plays a vital role in osteosarcoma treatment, malignant tumors, including osteosarcoma, exhibit strong chemoresistance, which is one of the primary reasons for their difficulty in eradication, easy recurrence, metastasis, and invasion [5]. Tumor aggressiveness, chemoresistance, and poor prognosis are often intertwined with the hypoxic state of the tumor microenvironment [6,7]. Hypoxia can influence tumor cell behavior through various mechanisms, including promoting proliferation, invasion, and metastasis, as well as inhibiting apoptosis [[8], [9], [10], [11]]. Hypoxia fosters tumor invasion and metastasis by initiating epithelial-mesenchymal transition (EMT) in tumor cells [12]. EMT is a critical biological process characterized by morphological and functional changes that equip tumor cells with increased migratory and invasive capacities [13,14]. As a solid tumor, osteosarcoma has been confirmed to harbor a hypoxic microenvironment, which is closely associated with chemoresistance. However, the precise mechanisms underlying hypoxia-induced chemoresistance in osteosarcoma remain incompletely understood [15,16]. Hypoxia-inducible factors (HIFs) are essential for regulating the tumor cell response to low oxygen conditions. The HIF family comprises three members, HIF-1, HIF-2, and HIF-3, each consisting of α and β subunits. The α subunit, whose expression is tightly controlled by intracellular oxygen levels, plays a pivotal role in HIF function. HIF-1α and HIF-2α are generally absent in most normal tissues but are highly expressed in diverse tumor cell types [17]. HIF-1α promotes blood supply, oxygen delivery, and energy provision by inducing VEGF and glycolytic enzyme expression, thereby ameliorating hypoxia [18]. It also upregulates MMP-2, MMP-7, and MMP-9 expression to facilitate malignant tumor metastasis [19]. Additionally, HIF-1α promotes malignant tumor metastasis by inducing EMT-related proteins such as SNAIL, TWIST, SLUG, ZEB1, SIP1, LOX, Vimentin, and N-cadherin [20]. Osteosarcoma tumors with high HIF expression are more likely to be aggressive and have a worse prognosis [[21], [22], [23]]. Prior research on hypoxia in osteosarcoma has primarily relied on bulk samples, which provide an average representation of the tumor [24,25]. The emergence of single-cell RNA sequencing (scRNA-seq) technology has introduced a novel approach to studying tumor heterogeneity and the microenvironment [26]. scRNA-seq enables the analysis of gene expression profiles of individual tumor cells and other cells within the tumor microenvironment, uncovering the functions and interactions of distinct cell subpopulations [27]. To explore the impact of hypoxia on osteosarcoma cells, specifically its role in mediating tumor invasion, we initially employed a cellular hypoxia prediction framework known as CHPF to predict the hypoxic status of individual cells. We then applied CHPF to single-cell RNA sequencing data obtained from osteosarcoma. Our study presents a detailed overview of cellular hypoxia within osteosarcoma and uncovers crucial regulatory factors within hypoxic subpopulations of osteosarcoma cells that demonstrate high invasive potential. These key regulators offer promising avenues for predicting patient survival and serve as potential therapeutic targets for treatments specifically targeting hypoxia in osteosarcoma. Methods Obtaining and analyzing transcriptome data Our predictive model was based on RNA expression and clinical data from 85 osteosarcoma cases in the TARGET database Additionally, an independent validation cohort was utilized to assess the model's stability and accuracy. All data were converted to TPM format and underwent log2 transformation for subsequent analysis. Furthermore, we integrated chip data from 53 osteosarcoma samples in the GSE21257 dataset from the GEO database as a validation set. Limma's normalize between arrays function was used to normalize the chip data. Single-cell RNA-sequencing data acquisition and processing We obtained single-cell RNA-sequencing (scRNA-seq) data for six osteosarcoma samples (two primary, two recurrent, and two metastatic) from the GEO database (accession number GSE152048). Analysis was performed using R version 4.1.3 and the Seurat package. To ensure data quality, we filtered out cells with high mitochondrial content (>20%) or high hematopoietic cell content (>3%). Additionally, we excluded cells with extremely low or high numbers of unique molecular identifiers (UMIs; 200–40,000) or genes (200–7000). The Seurat package's NormalizeData, FindVariableFeatures, and ScaleData functions were used for data normalization, selection of 2000 highly variable genes for analysis, and data transformation to mitigate cell cycle effects. We further corrected for batch effects using Harmony. Finally, we employed UMAP and TSNE for dimensionality reduction, allowing us to visualize the complex data in two or three dimensions. Louvain clustering was used to identify distinct cell populations within the scRNA-seq data. Differentially expressed genes between clusters were identified using the FindAllMarkers function with a significance threshold (p-value < 0.05), a minimum fold change threshold (log2 fold change > 0.25), and a minimum expression proportion threshold (>0.1). Identification of hypoxic cells The cellular hypoxia prediction framework, developed by Zhang et al., is a data-driven computational framework that integrates single-cell RNA sequencing transcriptomic data with established hypoxia gene signatures to predict the hypoxia status of individual cells at single-cell resolution in cancer [28]. Through construction of cell status-specific gene regulatory networks, CHPF identified 14 critical transcription factors driving hypoxia-induced tumor invasion, two of which were experimentally validated via knockdown assays. In our study, we utilized the CHPF software to predict cellular hypoxia based on single-cell transcriptome data and hypoxia gene signatures. Cell annotation analysis We annotated cells using specific markers: COL1A1, CDH11, and RUNX2 for osteoblastic cells; TOP2, PCNA, and MKI67 for proliferating osteoblastic cells; ACAN, COL2A1, and SOX9 for chondroblastic cells; CTSK and MMP9 for osteoclastic cells; IL7R, CD3D, and NKG7 for T and NK cells (TILs); CD74, CD14, and FCGR3A for myeloid cells; COL1A1, LUM, and DCN for fibroblasts; ACTA2 and RGS5 for pericytes; CXCL12, SFRP2, and MME (CD10) for MSCs; MYLPF and MYL1 for myoblasts; and PECAM1 and VWF for endothelial cells. Based on these annotations, we generated various visualizations, including TSNE plots and violin plots of cell markers. WGCNA analysis and enrichment analysis We employed the WGCNA package to analyze gene modules associated with specific cell populations and used the clusterProfiler package alongside GOBP and KEGG databases for gene enrichment analysis. Enrichment results were visually presented using EnrichmentMap and AutoAnnotate plugins in Cytoscape software. Analysis of cell-cell communication To evaluate potential cell-to-cell communications, we leveraged the CellChat tool. Initially, the normalized gene expression matrix was imported into CellChat to establish a CellChat object. Data preprocessing was conducted through the utilization of the built-in functions: identifyOverExpressedGenes, identifyOverExpressedInteractions, and ProjectData, all executed with their default parameters. Subsequently, the computeCommunProb, filterCommunication, and computeCommunProbPathway functions were applied, enabling the identification of potential ligand-receptor interactions. Ultimately, the aggregateNet function was employed to construct the comprehensive cell communication network. Single-cell CNV analysis We employed the infercnv package to predict copy number variations (CNVs) in tumor cells, specifically focusing on osteoblastic and chondroblastic subtypes, with endothelial cells serving as a reference point for comparison. CNV scores were calculated for each tumor cell. Single-cell transcription factor analysis The SCENIC package was employed to predict transcription factors in H3 and N5 cell populations. GRNboost2 was used for gene co-expression analysis to construct gene regulatory networks. To identify key players in the cell communication network, we evaluated the degree of connectivity between nodes. The top 1% of genes or transcription factors with the highest degree of connectivity were extracted for subsequent analysis. Construction of a prognostic model To identify prognostic markers within the H3 cell population, we analyzed unique and significant key hypoxia associated genes. Survival-related genes were identified using univariate Cox analysis. Subsequently, the LassoCox algorithm was employed to develop a prognostic model, and risk scores were calculated for each patient. High- and low-risk groups were defined using the median risk score. Predicting the response to immunotherapy We gathered datasets from a variety of cancer types, including low-grade glioma, lung squamous cell carcinoma, and melanoma, to predict immunotherapy response. GAS scores were calculated for each dataset to assess the potential efficacy of immunotherapy Cell culture, hypoxia treatment, and drug inhibition Osteosarcoma cells were cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin‑streptomycin at 37 °C in a humidified atmosphere containing 5% CO₂. To simulate the hypoxic tumor microenvironment, cells were placed in a modular incubator chamber and flushed with a gas mixture of 1% O₂, 5% CO₂, and 94% N₂, maintaining an oxygen concentration of 1‑2% for the indicated durations. For HIF‑1α inhibition, cells were treated with 1 µM of a specific HIF‑1α inhibitor (HY-115,903, MCE) for 24 h prior to and during hypoxia exposure. Western blot analysis Total protein was extracted using RIPA lysis buffer containing protease and phosphatase inhibitors. Equal amounts of protein were separated by SDS‑PAGE and transferred onto PVDF membranes. Membranes were blocked with 5% non‑fat milk and then incubated overnight at 4 °C with primary antibodies against SERPINH1, N‑cadherin, vimentin, HIF‑1α, and ACTIN (used as a loading control). After washing, membranes were incubated with HRP‑conjugated secondary antibodies for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL). Cell proliferation assays (CCK8 and EdU) Cell proliferation was assessed using the CCK8 assay. Cells were seeded into 96‑well plates at a density of 3 × 10³ cells per well. At each time point, 10 µL of CCK8 solution was added to each well and incubated for 2 h. Absorbance was measured at 450 nm using a microplate reader. For EdU incorporation assays, cells were incubated with 10 µM EdU for 2 h, then fixed with 4% paraformaldehyde and stained using a Click‑iT EdU kit according to the manufacturer’s instructions. Nuclei were counterstained with Hoechst 33,342. EdU‑positive cells were visualized and counted under a fluorescence microscope. Transwell invasion assay Cell invasion ability was evaluated using Transwell chambers (8‑µm pore size) pre‑coated with Matrigel (diluted 1:8 in serum‑free medium). Cells (1 × 10⁵) suspended in 200 µL of serum‑free medium were seeded into the upper chamber. The lower chamber was filled with 600 µL of complete medium containing 10% FBS as a chemoattractant. After 24 h of incubation under normoxic or hypoxic conditions, non‑invading cells on the upper surface were removed with a cotton swab. Invading cells on the lower surface were fixed with methanol, stained with 0.1% crystal violet, and counted in five randomly selected fields under a light microscope. Small interfering RNA (siRNA) knockdown SERPINH1 knockdown was achieved using specific small interfering RNA (siRNA) targeting SERPINH1 (si‑SERPINH1) or a negative control siRNA (si‑NC). Cells were transfected with 50 nM of siRNA using Lipofectamine 3000 reagent according to the manufacturer’s instructions. After 48 h of transfection, cells were subjected to normoxic or hypoxic conditions for subsequent experiments. Knockdown efficiency was confirmed by Western blot. Statistical analysis Throughout the entire process, from data processing and statistical analyses to visualizations, we relied solely on the R software, specifically version 4.1.3. To evaluate correlations among continuous variables, we utilized Pearson correlation coefficients. For comparing categorical variables, we opted for Chi-square tests. When dealing with continuous variables, we employed either Wilcoxon rank-sum tests or T-tests, depending on the data characteristics. Additionally, we leveraged the survival package to conduct Cox regression and Kaplan-Meier analyses. Results Characterization of the hypoxic landscape in osteosarcoma cells Dimensionality reduction and clustering of cells from osteosarcoma samples were performed using the UMAP algorithm, enabling precise identification of cell subpopulations through a series of cell type-specific marker genes. The distribution of cell subpopulations was presented through a UMAP plot, encompassing osteoblastic proliferative, chondroblastic, endothelial, osteoblastic, pericytic, myoblastic, T-cell, osteoclastic, and myeloid cells (Fig. 1A). Utilizing the CHPF software, cell hypoxia status was classified, revealing that hypoxic cells were primarily enriched in chondroblastic and osteoblastic subpopulations, with varying proportions across different samples and tissue types (Fig. 1B-C). A Sankey diagram offered additional insights into the intricate relationships between cell types, sample types, and hypoxia status. It corroborated the observation that hypoxic cells were primarily sourced from within the tumor mass, underscoring the reliability of our analytical approach(Fig. 1D). Furthermore, a heatmap visualization was utilized to illustrate the distinct expression patterns of hypoxia-related marker genes across both tumor and immune cell populations (Fig. 1E). Single-Cell Landscape of Osteosarcoma. A. UMAP plot of single-cell data colored by cell types. B. UMAP plot identifying hypoxic and non-hypoxic cells using CHPF software. C. Bar chart representing the proportion of hypoxic and non-hypoxic cells across samples, tissue types, and cell types. D. Sankey diagram shows the intricate relationships between different cell types, sample types, and their corresponding hypoxia status. E. A heatmap clearly differentiates between hypoxic and non-hypoxic states by displaying the expression patterns of hypoxia-related markers in both tumor and immune cells. Immune cell subpopulation analysis Given the pivotal role of immune cells in the hypoxic microenvironment, this study specifically examined the unique features of immune cells under hypoxic conditions. Dimensionality reduction and clustering analysis depicted the UMAP plot of immune cell subpopulations, distinguishing hypoxic from non-hypoxic states. Immune cells comprised CD8+ NKT cells, B cells, monocytes and macrophages, with hypoxic cells predominantly found in macrophages and monocytes (Fig. 2A-C). Further analysis revealed that these hypoxic macrophages and monocytes were primarily derived from patients with recurrent osteosarcoma (Fig. 2D). Immune-tumor cell communication networks were identified using CellChat analysis (Supplementary Figure 1A). A bubble plot of ligand-receptor interactions (Fig. 2F) highlighted the significant role of the MIF signaling pathway in intercellular communication. Correspondingly, a heatmap illustrating MIF pathway communication across different cell types is presented in Supplementary Figure 1B Analysis of Immune Cell Subpopulations. A-C. UMAP plots of immune cell subpopulations, clustering, cell types, and oxygen deprivation levels. D. Sankey diagram depicting the relationship between cell types, sample types, and oxygen deprivation levels for immune cells. E-F. Investigation of the interaction network between immune and tumor cells, emphasizing the intricacies of their communication. G. Dot plot showcasing the expression of MIF-(CD74+CXCR4) in single-cell data. Tumor cell subpopulation analysis Osteoblastic and chondroblastic cells were categorized as tumor cells and clustered based on hypoxia status. Hypoxic subpopulations were designated H1 to H4, while non-hypoxic subpopulations were named N1 to N5 (Fig. 3A-B). Hypoxia scores revealed the degree of hypoxia within each subpopulation, with H1 exhibiting the highest hypoxia score (Fig. 3C). Heatmap analysis highlighted that genes such as FGFBP2, S100A1, COL2A1, HAPLN1, and SNORC exhibited significant overexpression in the H1 subpopulation. These genes are intricately associated with cellular activities like migration, invasion, proliferation, and neovascularization, which collectively contribute to the aggressive phenotype of this subpopulation (Fig. 3D). Further examination via GOBP annotation and Cytoscape visualization illuminated the functional significance of hypoxia-associated markers, highlighting their roles in cellular activities such as mitosis, cell cohesion, and actin filament assembly (Fig. 3E). WGCNA analysis identified key gene modules correlated with each subpopulation, including the high correlation between the H4 subpopulation and the Mered module and between the N3 subpopulation and the MEturquoise module (Fig. 3F-G). An in-depth analysis examined the interplay between oxygen deprivation levels and cell maturation processes, uncovering that the H4 subpopulation exhibited the most advanced level of differentiation (refer to Fig. 3H). Furthermore, a progression analysis illuminated the positioning of tumor cells with heightened hypoxia scores towards the conclusion of the simulated temporal progression (as visualized in Fig. 3I). Supplementary Figure 2 provides UMAP plots and boxplots of tumor cell subpopulation distribution, hypoxia scores, and hypoxia subpopulation scores. Analysis of Tumor Cell Subpopulations. A-C. UMAP plots of hypoxic and non-hypoxic subpopulations, along with a ridge plot displaying hypoxia scores for each cell cluster. D. Visualization of gene expression markers across diverse cellular clusters through a heatmap representation. E. GOBP enrichment analysis results for each cell cluster. F. Heatmap illustrating the correlation between gene modules identified by WGCNA and each cell cluster. G. KEGG enrichment analysis results for the "red" and "turquoise" gene modules. H. CytoTRACE analysis results for each cell cluster. I. Pseudotime trajectory analysis of cell clusters using monocle3. Analysis of tumor-related pathways, CNVs, and transcription factors To uncover the correlation between hypoxia and osteosarcoma aggressiveness, we sourced genes pertinent to hypoxia, invasiveness, cell death, blood vessel formation, and EMT from the CancerSEA database. Subsequently, we employed GSVA to compute activity scores for each cellular subpopulation, providing insights into their functional profiles (Supplementary Figure 3). Results showed positive correlations between the hypoxia score and these biological processes (Fig. 4A). Among the various subpopulations, the H3 subpopulation demonstrated the highest degree of invasive potential, whereas the N5 subpopulation displayed the lowest (as illustrated in Fig. 4B). CNV analysis revealed copy number variations among tumor subpopulations (Supplementary Figure 5), and endothelial cells were used as a reference to assess CNV status in tumor cells (Fig. 4C). Transcription factor analysis further identified key genes in the H3 and N5 subpopulations, with their distribution illustrated through Venn diagrams (Fig. 4D-E). In the H3 subpopulation, 112 key genes were identified, of which 108 key genes were exclusive to the H2 subpopulation. Correlation analysis between H3-specific genes and Hallmark pathways revealed significant associations (|Cor|>0.3, P < 0.05), with FBOX5, POMC, and RPL32 positively regulating most markers (Fig. 4F). Analysis of Tumor-Related Pathways, CNVs, and Transcription Factors. A. A scatter graph depicting the relationship between oxygen deprivation index and four distinct signature scores. B. Boxplot comparing signature scores across cell clusters. C. CNV prediction results using infercnv software, with endothelial cells as the normal cell reference. D. Network diagram of top 1% transcription factors in H3 and N5 cell clusters. E. Venn diagram comparing top 1% transcription factors and genes between H3 and N5 cell clusters. F. Heatmap of correlations between H3-specific transcription factors and genes with Hallmark pathways (|Cor|>0.3, P < 0.05). Predicting osteosarcoma prognosis with H3-specific key genes Univariate Cox analysis was initially employed to screen for prognostic genes, followed by stepwise regression analysis to select OS-related genes, resulting in the identification of eight key genes (RPS28, ZC3HAV1, RPS8, S100A13, SERPINH1, EFNA5, RPL37A, USP11) for the final prognostic model (Supplementary Figure 4 C-D). The LassoCox algorithm refined the development of the prognostic model (as depicted in Supplementary Figure 4 A-B), enabling the stratification of patients into high- and low-risk categories based on an optimal cut-off point, rather than simply relying on the median value. A thorough correlation analysis between the expression levels of the model's genes and clinical data unveiled a stark contrast in survival rates, with patients in the low-risk group exhibiting significantly higher survival rates compared to those in the high-risk group (Fig. 5A-B). Multivariate Cox regression analysis reinforced the independent predictive value of these key gene signatures for OS with an odds ratio (OR) of 4.05 (95% confidence interval: 2.061–8.0, P < 0.001), as depicted in Fig. 5C. The prognostic potential of the eight key gene signatures was further validated using an external dataset (GSE21257), demonstrating statistically significant divergence in Kaplan-Meier curves between the two risk groups (P = 0.049), as illustrated in Fig. 5D. Prognostic Model Based on H3-Specific Transcription Factors and Genes. A. Heatmap of expression levels of model genes along with clinical indicators such as age. B. Survival analysis results for the TARGET dataset. C. Forest plot of multivariate Cox analysis incorporating risk score and clinical indicators. D. Survival analysis results for the external dataset GSE21257. Potential drug analysis for hypoxia-targeted therapy in osteosarcoma Based on GDSCv2 database information, correlations between H3-specific key genes and drugs were calculated, screening for key genes significantly associated with at least one compound (Fig. 6A). These compounds targeted various biological processes, including DNA replication, apoptosis regulation, cell cycle, and PI3K/MTOR signaling pathways (Fig. 6B). Notably, the RPS28 gene was found to be significantly associated with multiple drugs, including PRIMA-1MET (targeting p53 signaling), Gemcitabine (affecting DNA replication), Vinorelbine (interfering with mitosis), and Obatoclax Mesylate (regulating apoptosis) (Fig. 6C). To identify potential candidates for adjuvant hypoxia-targeted therapeutic interventions, we leveraged the Connectivity Map (CMap) database through the QUERY platform (clue.io). H3-specific key genes were submitted as upregulated, while N5-specific transcription factors and genes were submitted as downregulated. The resulting volcano plot depicted compound scores across cell lines (Fig. 6D). The top 5 compounds ranked by |Score| in each cell line were selected, totaling 43 compounds (Fig. 6E). Finally, the targeted pathways of these compounds were summarized (Fig. 6F), identifying promising candidates for future hypoxia-targeted therapies. Drug Analysis. A. Bar chart of statistically significant correlations between H3-specific transcription factors and genes with drugs. B. Bar chart of target pathways associated with drugs. C. Network diagram of the model gene RPS28, associated drugs, and their target pathways. D. Volcano plot of compound scores from CMap online analysis across cell lines. E. Bubble plot of top 5 compounds ranked by |Score| in each cell line. F. Bubble plot of target pathways corresponding to the top 5 compounds in each cell line. Hypoxia-induced changes in SERPINH1 expression, proliferation, and invasion in osteosarcoma cells To investigate the biological behavior of osteosarcoma cells under hypoxic conditions, we simulated the tumor hypoxic microenvironment using nitrogen to maintain an oxygen supply of 1–2%. First, we examined the expression of SERPINH1, one of the eight-gene prognostic signature identified in our study. Under hypoxia, SERPINH1 protein expression was significantly upregulated (Fig. 7A). Notably, treatment with a HIF-1α inhibitor markedly reversed this hypoxia-induced elevation. The expression pattern of SERPINH1 was consistent with that of epithelial-mesenchymal transition (EMT) markers, including N-cadherin and vimentin, which also increased under hypoxia and decreased upon HIF-1α inhibition. We further assessed cell proliferation using CCK8 and EdU incorporation assays. Hypoxia substantially increased the proliferative capacity of osteosarcoma cells, as evidenced by both CCK8 (Fig. 7B) and EdU labeling (Fig. 7C). This hypoxia-driven proliferative boost was significantly abrogated by HIF-1α inhibitor treatment. We then examined whether hypoxia could enhance the invasive phenotype of osteosarcoma cells, given the observed upregulation of EMT markers. Hypoxia indeed induced a significant increase in cell invasiveness, and this effect was closely correlated with the expression level of HIF-1α (Fig. 7D). Hypoxia-induced SERPINH1 expression promotes osteosarcoma proliferation and invasion. A. Protein levels of SERPINH1,—N-cadherin, and vimentin under normoxia, hypoxia, and hypoxia with HIF-1α inhibitor. B-C. CCK8 and EdU assays demonstrate hypoxia-induced proliferation is reversed by HIF-1α inhibition. D. Transwell assay shows hypoxia enhances invasiveness in a HIF-1α-dependent manner. E-F. SERPINH1 knockdown abolishes hypoxia-driven proliferation. G. SERPINH1 knockdown reverses hypoxia-induced invasion. To determine the functional role of SERPINH1 in hypoxia-induced aggressive behavior, we performed SERPINH1 knockdown experiments. Under normoxic conditions, SERPINH1 knockdown alone reduced the proliferative rate of tumor cells. More importantly, under hypoxic conditions, SERPINH1 knockdown completely abolished the hypoxia-induced increase in proliferation, bringing the proliferation rate back to the control level (Fig. 7E). These findings were further confirmed by EdU staining (Fig. 7F). In addition, Transwell invasion assays demonstrated that SERPINH1 knockdown similarly reversed the enhanced invasiveness induced by hypoxia (Fig. 7G). Taken together, these experimental results indicate that hypoxia promotes osteosarcoma cell proliferation and invasion through a HIF-1α/SERPINH1 axis, and that targeting SERPINH1 can effectively counteract hypoxia-induced malignant progression. Discussion In this study, we successfully identified various cell subpopulations, including osteoblastic proliferative, chondroblastic, and endothelial cells, through dimensionality reduction and clustering analysis of osteosarcoma samples using the UMAP algorithm. This identification of cell subpopulations provides critical insights into the cellular heterogeneity of osteosarcoma. By applying the CHPF software, we further explored the distribution of hypoxia status across different cell subpopulations, revealing that hypoxic cells were primarily enriched in chondroblastic and osteoblastic subpopulations. This discovery aligns with previous research endeavors, which have underscored the pivotal role played by the hypoxic microenvironment in both the development and progression of osteosarcoma [29,30]. The relationship between hypoxia status and the tumor microenvironment was further illuminated by the application of Sankey diagrams, confirming that hypoxic cells predominantly originate from the tumor core region. This study underscores the pivotal role of the hypoxic microenvironment, specifically within the tumor core, in osteosarcoma, where it contributes to tumor aggressiveness and therapeutic resistance through various mechanisms, consistent with previous research emphasizing its significance in the disease's development and progression [31,32]. Our research focused on the hypoxic signatures exhibited by immune cells residing within the tumor microenvironment, with a particular emphasis on the hypoxic condition of macrophages and monocytes. These cells, primarily sourced from patients experiencing recurrent osteosarcoma, imply that hypoxia might hold a pivotal influence in the recurrence of the disease, particularly in terms of its modulation of immune cell functionality. Hypoxia, a common phenomenon in the tumor microenvironment, promotes tumor growth and escape by influencing immune cell function [33]. Hypoxic environments can elicit an immunosuppressive phenotype via HIF-1-mediated pathways, leading to alterations in molecular markers, immune cell migration patterns, and angiogenesis [34]. Additionally, oxidative stress triggers a cascade of events, including lipid peroxidation, endoplasmic reticulum stress, and Treg dysfunction, all of which contribute to immune dysregulation [34]. This indicates that hypoxia not only directly affects immune cell function but also indirectly influences immune responses through metabolites. The potential communication networks between immune and tumor cells revealed by CellChat software, particularly the role of the MIF signaling pathway, offer new targets for future therapeutic strategies targeting immune cells. This aligns with the role of HIF in regulating immune cell function, as HIF is also a pivotal regulator of immune cell function [35]. Through cluster analysis of hypoxia status, we identified distinct hypoxic subpopulations and explored their biological characteristics. Genes such as FGFBP2 and S100A1 exhibited marked overexpression in hypoxic subpopulations, displaying strong correlations with cellular activities like proliferation, motility, aggressiveness, and neovascularization [36,37]. The predominant presence of these genes within signaling cascades linked to tumor cell proliferation, motility, and invasiveness serves as a theoretical foundation for the development of future therapeutic interventions aimed at these specific pathways. Our findings established a direct positive correlation between increasing hypoxia scores and heightened tumor aggressiveness, with the H3 subpopulation exhibiting the most pronounced association. This suggests that hypoxia may impact osteosarcoma prognosis by promoting tumor cell aggressiveness. Hypoxic conditions enhance tumor cell invasion and metastasis capabilities [38,39], and are associated with increased malignancy. Transcription factor analysis further identified key genes in the H3 and N5 subpopulations. F-box proteins, which serve as indispensable components within E3 ubiquitin ligase complexes, hold a pivotal position in regulating cell cycle progression and the onset of cancer development [40,41]. By recognizing and tagging specific proteins for degradation, F-box proteins maintain cellular homeostasis and prevent tumorigenesis [42]. Thus, FBOX5, belonging to the F-box protein superfamily, may participate in modulating metabolic processes and survival mechanisms in tumor cells, especially in response to hypoxic environments. These transcription factors play pivotal roles in regulating tumor cell biology and may represent important targets for future therapeutic strategies. Using information from the GDSCv2 database, we analyzed the correlations between unique transcription factors and genes in the H3 subpopulation and drugs, uncovering significant associations with RPS28 and various drugs. This provides new targets for future therapeutic strategies targeting these drugs. The RPS family, a crucial component of ribosomes involved in multiple stages of protein synthesis, has been implicated in both cellular basic physiological functions and tumor development, as well as its adaptation to hypoxic environments [[43], [45], [44], [45]]. We acknowledge that due to the limited number of immune cells captured in our scRNA-seq dataset and the overall low immune infiltration characteristic of osteosarcoma, we are unable to reliably assign M1-like or M2-like polarization states to the hypoxic macrophages identified in this study. More importantly, accumulating evidence suggests that the simple M1/M2 dichotomy is insufficient to capture the functional diversity and plasticity of tumor-associated macrophages in the hypoxic tumor microenvironment, and a more nuanced characterization is warranted in future investigations. We recognize several limitations of our prognostic model. First, due to the limited availability of osteosarcoma cohorts with both transcriptomic and survival data, our model was developed and validated on only two datasets (TARGET and GSE21257), which may not fully capture the heterogeneity of the disease. Second, the multi‑step selection procedure (univariate Cox followed by LassoCox) carries a potential risk of overfitting, and the model's generalizability should be further assessed in larger, independent prospective cohorts. Conclusion In conclusion, we employed CHPF as a tool to ascertain cellular hypoxia status and gain insights into the intratumoral variability of hypoxia within osteosarcoma. Notably, our investigation unveiled a unique hypoxic subpopulation characterized by heightened invasive capability, accompanied by distinctive transcriptomic and genomic profiles. Our findings indicate that RPS28, ZC3HAV1, RPS8, S100A13, SERPINH1, EFNA5, RPL37A, and USP11 are key hypoxia associated genes driving hypoxia-induced tumor aggressiveness, which can serve as independent prognostic markers for predicting osteosarcoma survival. Our discoveries offer profound insights into the molecular characteristics and mechanisms underlying hypoxia-associated osteosarcoma aggressiveness. Clinical trial number Not applicable Funding Not applicable. Consent to publish declaration All authors agree to publish Ethics declaration All data used in this study were obtained from publicly available repositories (TARGET, GEO). These datasets are de identified and were collected with appropriate informed consent under their respective institutional review board (IRB) approvals. The secondary analysis of de identified public data does not require additional IRB approval at our institution (Wenzhou Medical University), as confirmed by the Ethics Committee of The Second Affiliated Hospital and Yuying Children’s Hospital. Data availability declaration All data can be requested from the corresponding author CRediT authorship contribution statement Shangfang Xie: Writing – original draft, Data curation, Conceptualization. Wenyao Yu: Data curation, Conceptualization. Runye Lin: Investigation. Songjian Xin: Writing – review & editing, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments Thanks to all participants and staff for their contributions. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.tranon.2026.102863. Supplementary materials References Associated Data Supplementary Materials Data Availability Statement All data can be requested from the corresponding author