Entosis remodels the immune microenvironment of osteosarcoma by regulating macrophage polarization: An integrated study based on multi-omics data and in vitro experiments

Highlights

• •High Entosis activity predicts favorable prognosis in osteosarcoma.
• •Entosis remodels immune microenvironment by regulating macrophage polarization.
• •The Macro_CXCL2 subtype is the core target of Entosis regulation.
• •A six-gene Entosis-related model effectively predicts osteosarcoma survival.

Abstract

Background

Entosis is a non-apoptotic form of programmed cell death. Its clinical significance and mechanism of immune microenvironment remodeling in osteosarcoma remain poorly understood.

Methods

Multi-omics data from TARGET-OS and GEO were integrated with in vitro experiments. ssGSEA, WGCNA, scRNA-seq, pseudotime analysis, and cell-cell communication were used to assess Entosis activity and immune association. A prognostic risk model was constructed and validated. In vitro experiments (qPCR, flow cytometry, Western blot, ELISA) validated Entosis regulation of macrophage polarization.

Results

High Entosis activity was significantly correlated with favorable prognosis and a highly activated immune microenvironment in osteosarcoma patients. scRNA-seq identified macrophages as the core target cells of Entosis regulation, and the Macro_CXCL2 subset was characterized as the functional subpopulation with the highest Entosis score at the terminal differentiation stage. Cell communication analysis revealed that, under high Entosis conditions, macrophages enhanced crosstalk with other immune cells via the SPP1-CD44 and TNF signaling axes. A six-gene prognostic model (CD163, FCGR2A, NPC2, CD209, SAMHD1, IKZF1) demonstrated robust performance across multiple datasets. In vitro experiments further confirmed that high-Entosis tumor cells upregulated CXCL12 secretion in a paracrine manner, driving macrophages toward a specific functional phenotype. Mechanistically, Entosis activation was dependent on the ROCK signaling pathway and synergized with the hypoxia, TNFα-NFκB, and MAPK pathways to coordinately regulate malignant cell stemness and immune microenvironment remodeling.

Conclusion

Entosis exerts tumor-suppressive and immune-activating effects in osteosarcoma by remodeling macrophage function, offering new targets for precision prognosis and Entosis-directed immunotherapy.

Introduction

Osteosarcoma (OS) is a highly aggressive malignant bone tumor that mainly affects children under nine years of age, with an incidence rate of 2.3 cases per million population over the past decade [1]. This disease is common in long bones, especially around the knee joint, and manifests as severe pain, bone swelling, and in some cases pathological fractures [2]. About 25% of patients have metastatic lesions when diagnosed, with the lungs being the most common site of metastasis [3]. Although the comprehensive treatment plan of surgery combined with radiotherapy and chemotherapy has significantly improved the local control rate of patients in recent years, distant metastasis and treatment resistance are still the main reasons for the poor prognosis of patients. The 5-year survival rate is only 60%-70%, and the survival rate of patients with advanced stage or recurrence is less than 20% [4]. Therefore, in-depth analysis of the molecular mechanisms of cancer development and the discovery of biomarkers and potential therapeutic targets with prognostic predictive value are crucial to optimizing precise diagnosis and treatment strategies for cancer.

Cell death is a basic physiological process that regulates various physiological phenomena such as growth, development, aging and disease. Unlike accidental cell death (ACD), correlated cell death (RCD)—also known as programmed cell death (PCD)—can be triggered by specific signaling mechanisms and/or metabolic reprogramming [5,6]. Programmed cell death (PCD), also known as regulated cell death, plays a key role in maintaining tissue homeostasis and clearing damaged or abnormal cells (9). PCD includes ferroptosis, apoptosis, pyroptosis, autophagy, necroptosis, copper death, protoplast death, endocytic death, reticular death, lysosome-dependent death, alkaline death, oxidative death, reticular death, immunogenic death, extracellular matrix stripping-induced death, parapoptosis, methylation death, endocytic death, MPT-driven necrosis and disulfide bond death [[7], [8], [9], [10]]. Entosis is a cell phagocytosis process that relies on the actin cytoskeleton, in which a cell is engulfed by another adjacent cell and then dies within the phagocyte. Its role in tumor microenvironment remodeling, immune regulation, and tumor cell survival has gradually attracted attention [11]. Studies have shown that Entosis is abnormally expressed in various tumors such as liver cancer and breast cancer, and is related to patient prognosis and tumor malignancy [12]. However, the expression characteristics, regulatory mechanism and interaction between Entosis and the tumor microenvironment in osteosarcoma have not yet been systematically reported.

Based on this, this study integrated batch RNA sequencing and single-cell RNA sequencing (scRNA-seq) data to analyze the regulatory mechanism of Entosis on the osteosarcoma tumor microenvironment and malignant cell biological behavior, and constructed a prognostic prediction model based on Entosis-related genes, providing new theoretical basis and potential targets for prognostic assessment and individualized treatment of osteosarcoma patients.

Methods

Data collection

Download the standardized and log2-transformed TPM expression profiles and clinical information for the TARGET-OS cohort from the UCSC Xena database (https://xenabrowser.net/datapages/). We obtained the batch RNA sequencing dataset GSE21257, as well as scRNA-seq datasets GSE162454 and GSE152048, from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/). GSE162454 and GSE152048 were used for single-cell transcriptome analysis, while GSE21257 served as the external validation set for the risk model.

Genetset score

We collected Entosis-related genes [12], immune function gene sets [13], and 17 cell apoptosis gene sets [14,15] from existing literature. Subsequently, we employed the single-sample gene set enrichment analysis (ssGSEA) algorithm within the R package “GSVA” to calculate the enrichment scores for each specific gene set across all samples in the TARGET-OS cohort, based on its gene expression data.

Survival analysis

The Kaplan-Meier curve is an effective tool for survival analysis, capable of describing how patient survival rates change over time. In biomedical research, particularly in disease prognosis analysis, Kaplan-Meier curves are widely applied. Tumor samples are stratified into high- and low-risk groups using the median value of a specific characteristic or indicator as the threshold. Survival curves for prognosis analysis are generated via the Kaplan-Meier method, and the log-rank test is employed to determine the significance of differences. A p-value < 0.05 indicates a statistically significant survival difference between the two groups.

Differential expression analysis

Differential expression analysis of the TARGET-OS dataset was performed using the “limma” package in R software. Significant differentially expressed genes were identified based on the criteria |log2FC| > 0.25 and adjusted_pvalue < 0.05. Expression patterns of these genes were visualized as volcano plots and heatmaps using the “ggplot2” and “pheatmap” packages in R, respectively. Gene set enrichment analysis (GSEA) enrichment analysis was performed on overlapping differentially expressed genes (DEGs) using the ClusterProfiler package. Analysis parameters were set with minGSSize = 3 and pvalueCutoff = 0.1. The top 10 most significant gene sets from the enrichment results were selected and visualized using line charts.

ScRNA-seq data processing and Entosis score calculation

ScRNA-seq data preprocessing was performed using the “Seurat” [16,17] package in R software. The “PercentageFeatureSet” function was employed to assess the proportion of mitochondrial genes in the dataset. To ensure data quality and integrity, only genes expressed in at least three cells were retained; Additionally, each cell had to meet the criteria of expressing between 200 and 10,000 genes, with mitochondrial genes constituting less than 20% of the total. Following preprocessing, the scRNA-seq data were normalized using the “NormalizeData” function. The normalized data were converted into a Seurat object, and 2,000 highly variable genes were selected via the “FindVariableFeatures” function. Subsequently, the “RunPCA” tool performed scaling and principal component analysis on this set of highly variable genes. Dimension reduction and visualization in a two-dimensional coordinate space were achieved using the Shared Nearest Neighbor (SNN) modular optimization algorithm and the t-distributed Stochastic Neighborhood Embedding (t-SNE) clustering algorithm. To identify specific marker genes for each cell cluster, we performed Wilcoxon tests on different cell types using the “FindAllMarkers” and ‘FindMarkers’ algorithms, followed by annotation of various cell subpopulations based on these marker genes. For evaluating entosis scores in osteosarcoma single-cell sequencing data, we employed three widely used algorithms: “AUCell,” “singscore,” and “UCell.” Specifically, “AUCell” is an R-language tool for evaluating the enrichment level of specific gene sets within individual sample gene expression data, aiding in the analysis of biological processes and disease mechanisms; the ‘UCell’ method is suitable for unsupervised cell type identification, enabling classification of specific cell types without relying on historical information or labels; “singscore” is used to quantify the activity level of specific biological functions or processes within individual samples or cells, serving as an effective means for assessing cellular states.

Pseudotime analysis

An analysis of pseudotime, also known as cell trajectory analysis, helps predict the evolutionary trajectory of apoptosis pathways and cell subtypes and infer the differentiation trajectory of stemcells during disease progression. By analyzing key gene expression patterns using Monocle 2, we performed pseudotime analysis in the current study. The pseudotime value was used by Monocle to model the gene expression level as a nonlinear smooth pseudotime function to show changes in gene expression with time. FDR <1e−5 was regarded as a significant difference.

Cell–cell interaction analysis

Based on the ligand–receptor information, we used the single-cell gene expression matrix to unravel the communication between immune cell subtypes which was contained in CellChat software (http://www.cellchat.org/) with default parameters, modeling the communication probability and identifying significant communications.

Copy number alteration inference

The CNV was evaluated by “infercnv” R package (version 1.12.0) to differentiate between malignant and non-malignant cells. The different genomic locations of CNVs were presented.

Construction of weighted gene co-expression network analysis (WGCNA)

WGCNA is a powerful analytical technique capable of extracting significant insights from high-throughput gene expression data, thereby enabling the analysis of gene regulatory network structure and function [18]. Using the WGCNA package in R software, WGCNA analysis was performed on the top 25% most variable genes in tumor samples. Anomalous samples were detected and excluded using sample-level clustering to select an appropriate soft threshold power for achieving a standard scale-free network. The subsequent phase involved constructing adjacency and topological overlap matrices (TOM), calculating phase differences, and employing dynamic tree cutting to generate gene dendrograms and identify modules with a minimum size of 40. Highly similar modules were then merged through clustering of module signature genes, enabling the calculation of correlations between these signature genes and Entosis activity. Finally, the most highly correlated modules were selected for further analysis.

Construction of risk scoring models

Univariate Cox regression analysis using the “survival” package identified statistically significant prognostic genes (p < 0.05). Subsequently, Lasso and multivariate stepwise regression analysis [19] were employed to further select genes and risk factors highly correlated with prognosis. Based on coefficients derived from multivariate analysis, a risk score was assigned to each osteosarcoma patient. The risk score calculation formula is as follows:

$$\text{Riskscore}=\sum _{\mathrm{i}=0}^{\mathrm{n}}\beta \mathrm{i}*\chi \mathrm{i}$$

Where β_i represents the weight coefficient of each gene in the model; χ_i represents the expression level of the corresponding gene.

Additionally, Kaplan-Meier methods were employed to construct survival curves for predicting the prognosis of osteosarcoma patients, with log-rank tests used to determine the significance of differences. The predictive model's effectiveness was evaluated using receiver operating characteristic (ROC) curves [20], where an area under the curve (AUC) value exceeding 0.65 was considered indicative of good model performance. Simultaneously, the R package “ggplot2” was utilized to generate scatter plots of survival time versus survival status.

Unsupervised clustering of model genes

Based on the model genes, we performed unsupervised clustering using the “ConsensusClusterPlus” software package to delineate molecular subtypes. The parameters “maxK” was set to “10”, “clusterAlg” was set to “km”, and ‘distance’ was set to “pearson”.

Cell culture

Mouse osteosarcoma cell lines (DUNN) and BMDM were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. Cells were maintained in a humidified incubator at 37°C with 5% CO₂. For suspension culture, cells were seeded in ultra-low attachment plates to induce Entosis. For inhibitor treatment, cells were pretreated with ROCK inhibitor Y-27632 (Beyotime, SC0326, China) for the indicated duration. Conditioned medium was collected after centrifugation and filtered prior to macrophage stimulation.

Western blot analysis

Tissues were homogenized in RIPA lysis buffer containing protease and phosphatase inhibitors. Total protein levels were quantified using a BCA assay. Equal amounts of protein (20 μg per sample) were separated by SDS‑PAGE and transferred to PVDF membranes. Membranes were blocked with 5% non‑fat milk in PBST for 1 h at room temperature, then incubated overnight with primary antibodies at 4°C. The primary antibodies used in this study are listed below:

• Anti-p38 (Proteintech, 14064-1-AP, USA); Anti-p-p38 (Proteintech, 28796-1-AP, USA); Anti-ERK1/2(Proteintech, 11257-1-AP); p-ERK1/2 (Proteintech, 80031-1-RR, USA); Anti-p-p65 Polyclonal Antibody (Proteintech, 82335-1-RR, USA); Anti- p65 Polyclonal Antibody (Proteintech, 80979-1-RR, USA); Anti-HIF-1α Polyclonal Antibody (Proteintech, 20960-1-AP, USA); Anti-LAMP1 Polyclonal Antibody (Proteintech, 21997-1-AP USA); Anti-LC3 Polyclonal Antibody (Proteintech, 14600-1-AP, USA); Anti-Alpha Tubulin Polyclonal Antibody (Proteintech, 80762-1-RR, USA). After extensive PBST washing, membranes were incubated with HRP-linked secondary antibodies for 1 h at room temperature. Protein bands were visualized via enhanced chemiluminescence (ECL), and band intensities were quantified using ImageJ software. Anti-α-Tubulin polyclonal antibody (Proteintech, 80762-1-RR, USA) was used as the internal reference.

Fluorescence-activated cell sorting (FACS)

Single-cell suspensions were washed with FACS buffer. The cells were incubated with the indicated antibodies for 30 min at 4°C, followed by fixation with 4% paraformaldehyde. Flow cytometry data were acquired using a FACSCanto flow cytometer (BD Biosciences, USA) and analyzed with FlowJo software (Tree Star, USA).

mRNA purification and reverse transcription quantitative PCR (RT-qPCR)

Total RNA was isolated from cells via TRIzol (Sigma-Aldrich, T9424, USA). cDNA was synthesized using a Yeasen cDNA kit (Yeasen, 11119ES60, China), and qPCR was conducted on an ABI system with SYBR Green Master Mix (Yeasen, 11184ES03, China). Expression levels were calculated by the 2^-ΔΔCt method (normalized to GAPDH), with all samples tested in triplicate. See Table 1 for primer sequences.

Table 1.

Primer sequences.

Primer name	Primer sequences
Mouse GAPDH Forward	AGCTTGTCATCAACGGGAAG
Mouse GAPDH Reverse	TTTGATGTTAGTGGGGTCTCG
Mouse iNOS Forward	GTTCTCAGCCCAACAATACAAGA
Mouse iNOS Reverse	CGCAGCTCTAGGAGCATGTG
Mouse TNF-α Forward	CATCTTCTCAAAATTCGAGTGACAA
Mouse TNF-α Reverse	TGGGAGTAGACAAGGTACAACCC
Mouse IL-1β Forward	CCGTGGACCTTCCAGGATGA
Mouse IL-1β Reverse	GGGAACGTCACACACCAGCA
Mouse ARG1 Forward	GTGAAGAACCCGGAGACCTG
Mouse ARG1 Reverse	GGTGATGCCAGAGATGCTTCC
Mouse CD206 Forward	TCTTTGCCTTTCCCAGCTTC
Mouse CD206 Reverse	TGGAAGTCATGGTTCCGTGT
Mouse CXCL12 Forward	CGCTCTGCATCAGTGACGGTA
Mouse CXCL12 Reverse	GTTCTTCAGCCGTGCAACAATC
Mouse CD163 Forward	TACAATGGAGCTTGGGGCAG
Mouse CD163 Reverse	TGCCTCGATGGTGTCTTGTC
Mouse FCGR2A Forward	GCTGCTGGAGAACCTGAAGA
Mouse FCGR2A Reverse	GGAGTTGATGGTGGTGATGG
Mouse NPC2 Forward	GCTGGATGTGGAGAAGGTCA
Mouse NPC2 Reverse	CCAGGTATGGTGACGATGCT
Mouse CD209 Forward	CAAGGAGACCTTCACCTTGG
Mouse CD209 Reverse	GGTAGTTGACGTGGTTGCTG
Mouse SAMHD1 Forward	AAAACCAGGTTTCACAACTTCTGC
Mouse SAMHD1 Reverse	TGCGGCATACAAACTCTTTCTGT
Mouse IKZF1 Forward	AATGAATGGCTCCCACAGGG
Mouse IKZF1 Reverse	GAACCATGAGCACATTGGGC

Enzyme-linked immunosorbent assay (ELISA)

The concentrations of CXCL12 in samples were quantified using commercially available ELISA kits (GOY-1904E, Goybio, China) according to the manufacturers’ instructions.

Statistical analysis

R version 4.1.3 was used for all statistical studies. The survival curve was plotted by the Kaplan–Meier survival curve and the P-value was obtained by the log-rank test. The Student’s t-test and Wilcoxon test were used to compare the differences between the two groups, and Spearman analysis was used to calculate the correlation coefficients. Double-tailed P < 0.050 was considered statistically significant.

Results

Enrichment features of Entosis-related gene sets

We obtained 85 osteosarcoma samples with clinical information from the UCSC Xena database. Using the ssGSEA method, we performed single-sample gene set enrichment analysis on the Entosis gene set to obtain enrichment scores. Based on the median score, the cohort was divided into high-score and low-score groups, followed by Kaplan-Meier survival curve analysis (Fig. 1A). Results revealed a significant difference in survival prognosis between the two groups (P < 0.05), with patients in the Entosis-high group demonstrating markedly superior survival outcomes compared to the Entosis-low group. Heatmap analysis indicated significant differences in 25 immune functions between the Entosis-high and Entosis-low groups, with significantly higher enrichment levels of immune function-related genes observed in the Entosis-high group (Fig. 1B). Furthermore, we assessed differences in enrichment scores for 17 cell death pathways between groups. The Entosis-high group showed significantly elevated enrichment levels for cell death-related gene sets including hypoxia-related genes, ferroptosis, lysosome-dependent cell death, immunogenic cell death, and necrotic apoptosis (Fig. 1C).

Fig. 1.

Analysis of Entosis-related gene sets and immune-relatedness in osteosarcoma. (A) Kaplan-Meier curve analysis comparing Entosis-high and Entosis-low groups. (B) Heatmap of 25 differentially expressed immune functions. (C) Heatmap of enrichment scores for 17 differentially expressed cell death pathways. (D) Volcano plot showing gene distribution differences between Entosis-high and Entosis-low groups. (E) Enrichment analysis results for immune function-related gene sets among differentially expressed genes. (F) HALLMARK gene set enrichment analysis results for differentially expressed genes. (G) KEGG pathway enrichment analysis results for differentially expressed genes. * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001.

Subsequently, the “limma” software package was used to screen for DEGs between the two groups (Fig. 1D), identifying a total of 632 significantly differentially expressed genes, including 611 upregulated genes and 21 downregulated genes. Finally, GSEA was performed on the highly expressed genes in the Entosis-high group to explore their associated biological processes and functional pathways. GSEA analysis of immune function-related gene sets showed enrichment in antigen processing and presentation, chemokine ontology, CSF1 response (mediating macrophage recruitment), and MHC-II molecule-related gene sets (Fig. 1E). GSEA analysis of HALLMARK gene sets revealed enrichment in allogeneic transplant rejection, epithelial-mesenchymal transition, inflammatory response, and KRAS signaling pathway activation-related gene sets (Fig. 1F); KEGG GSEA analysis showed enrichment in pathways related to antigen processing and presentation, asthma, hematopoietic cell lineage, IgA production in the intestinal immune network, leishmaniasis, Staphylococcus aureus infection, systemic lupus erythematosus, Th17 cell differentiation, tuberculosis, and viral myocarditis (Fig. 1G). Collectively, these findings suggest a highly active tumor immune microenvironment in Entosis-high osteosarcoma tissues, potentially conferring greater sensitivity to immune checkpoint inhibitor therapy or macrophage-targeted treatments (e.g., CSF1R inhibitors). This provides a potential biomarker for precise stratification and personalized treatment strategies in osteosarcoma patients.

Enrichment scores for Entosis-related genes in scRNA-seq

To reveal the dynamic changes in the tumor microenvironment (TME) during osteosarcoma development, we acquired two independent single-cell datasets from osteosarcoma samples. After rigorous quality control and data normalization, we obtained transcriptomic data from 95,598 cells. Following batch effect correction, UMAP dimensionality reduction and clustering visualized the cellular distribution patterns (Fig. 2A). Subsequently, standard cellular markers were used for annotation, identifying 12 major cell clusters: macrophages, endothelial cells, plasma cells, B cells, NK/T cells, fibroblasts, myofibroblasts, chondrocytes, monocytes, osteoclasts, osteoblasts, and mesenchymal stem cells (Fig. 2B). Bubble plot visualization successfully depicted the key marker gene expression profiles across cell types (Fig. 2C). Heatmaps further illustrated the expression distribution patterns of Entosis-related marker genes across cell subpopulations (Fig. 2D). Subsequently, based on Entosis-related gene expression data, we assessed Entosis activity levels in scRNA-seq data. Three widely used algorithms (UCell, AUCell, and singscore) were employed to score gene sets for Entosis activity (Fig. 2E). Results revealed significantly higher Entosis scores in macrophages and monocytes compared to other cell subpopulations. Cells were stratified into high- and low-Entosis-score groups using the median Entosis score across all cells as the cutoff. Bar chart analysis demonstrated a similarly elevated proportion of macrophages and monocytes in the high-score group (Fig. 2F). Further statistical analysis revealed a significant positive correlation between macrophages and high Entosis scores (Fig. 2G), indicating that macrophages are the cell type most closely associated with Entosis activity. This finding provides a key cellular target direction for subsequent investigations into the regulatory role of Entosis in tumor progression and the development of targeted intervention strategies.

Fig. 2.

Elevated Entosis scores in tumor macrophages. (A) UMAP plot showing cell distribution characteristics of batch-effect-corrected osteosarcoma single-cell transcriptome data. (B) Cell types identified by marker genes. (C) Bubble plot showing expression patterns of key marker genes across cell types. (D) Heatmap displaying expression patterns of entosis-related marker genes across cell types. (E) Bubble plot showing entosis activity scores across cell types. (F) Bar chart showing the proportion of cells with high vs. low entosis scores in each cell type. (G) Heatmap showing differences between high- and low-Entosis-score groups across cell types.

Identification of malignant cells

This study selected mesenchymal stem cells (MSCs), osteoclasts, and osteoblasts as research subjects, using NK/T cells as reference cells, and identified malignant cells through inferCNV analysis. Based on copy number variation (CNV) scores, all candidate cells were clustered into 10 cell clusters (Fig. 3 A); Violin plot results revealed that clusters 1, 2, 5, 6, 8, 9, and 10 exhibited significantly higher CNV levels than other clusters (Fig. 3B). These seven clusters were thus defined as malignant cell clusters. To elucidate the biological characteristics of malignant cell subpopulations, CytoTRACE2 and Entosis scores were further evaluated. UMAP visualization revealed substantial overlap between high CytoTRACE2-scoring and high Entosis-scoring regions (Fig. 3C, D). Bar chart analysis (Fig. 3E) indicated significant differences in cellular differentiation states between high- and low-Entosis-scoring subpopulations, with the high-Entosis group exhibiting a higher proportion of pluripotent cells. Further statistical analysis demonstrated a significant difference in CytoTRACE2 scores between high- and low-Entosis groups (Fig. 3F). Correlation analysis confirmed a significant positive correlation between Entosis and CytoTRACE2 scores (Fig. 3G), suggesting Entosis levels may regulate the differentiation potential of malignant cells and providing crucial insights into the mechanisms underpinning malignant stemness maintenance. Based on these findings, differential expression analysis and GSEA were performed on malignant cell subpopulations with high and low Entosis scores. Results revealed that the hypoxia pathway, tumor necrosis factor α (TNFα)-mediated nuclear factor κB (NFκB) signaling pathway, and mitogen-activated protein kinase (MAPK) signaling pathway were significantly activated in the high Entosis score group. In contrast, the low Entosis score group was characterized by high activation of the oxidative phosphorylation pathway and ribosome pathway (Fig. 3H, I). To validate the reliability of these pathway differences, bubble plots were used to visualize expression pattern disparities of hypoxia- and MAPK-related genes between the two subpopulations (Fig. 3J, K). Results demonstrated that key genes in these pathways were significantly overexpressed in the high Entosis score group, further corroborating the pathway activation characteristics of the high Entosis score malignant cell subpopulation.

Fig. 3.

Cell cluster identification and InferCNV analysis. (A) Cell clustering map based on CNV scores. (B) Violin plots of CNV scores across cell subpopulations. (C) UMAP annotation plot of malignant cells' CytoTRACE2 scores. (D) UMAP annotation plot of malignant cell Entosis scores. (E) Bar chart showing pluripotent cell proportions in high- and low-Entosis score groups. (F) Differential analysis of CytoTRACE2 scores between high- and low-Entosis score groups. (G) Scatter plot correlating Entosis scores with CytoTRACE2 scores. (H) GSEA enrichment results based on HALLMARK terms. (I) GSEA enrichment results based on KEGG terms. (J) Expression bubble plot of hypoxia pathway-related genes in high- and low-Entosis score groups. (K) Expression bubble plot of MAPK signaling pathway-related genes in high- and low-Entosis score groups. **** p<0.0001.

Identification and characterization of macrophages

The infiltration status and functional regulation of immune cells within the TME are critical factors influencing tumor progression. As a core immune cell type, the interaction between macrophages and Entosis, along with their biological functions, remain poorly understood. Therefore, this study further focuses on macrophages to explore their regulatory association with Entosis and functional characteristics. Based on Entosis scores, macrophages were classified into a high-Entosis-score subpopulation (EnhighMacro) and a low-Entosis-score subpopulation (EnlowMacro). Differential expression analysis and GSEA enrichment analysis were performed on both subpopulations. Results revealed that in GSEA analysis using HALLMARK gene sets, the EnhighMacro subpopulation showed significant enrichment in immune-related pathways including allograft rejection, inflammatory response, and interferon-gamma response (Fig. 4A). In analyses of immune function gene sets and KEGG gene sets, this subpopulation exhibited significant activation of chemokine/cytokine-related functions and inflammatory signaling pathways such as NF-κB and TNF (Fig. 4B, C), suggesting that the high Entosis score macrophage subpopulation may participate in TME remodeling by regulating inflammatory immune responses.

Fig. 4.

Identification of macrophage subtypes (A) GSEA based on HALLMARK gene sets. (B) GSEA based on immune function gene sets. (C) GSEA based on KEGG gene sets. (D) UMAP visualization of different macrophage subtypes. (E) Heatmap of the top 5 most significantly upregulated differentially expressed genes in each macrophage subtype. (F) Violin plot of Entosis scores across macrophage subtypes. (G) Heatmap of Entosis score correlation analysis with macrophage subtypes. (H) GO enrichment analysis of marker genes for the Macro_CXCL2 subtype. **** p<0.0001.

To further explore the intrinsic relationship between macrophages and entosis, UMAP analysis subdivided macrophages into seven distinct subtypes: Macro_APOCA, Macro_LUM, Macro_ISG15, Macro_MT1H, Macro_MKI67, Macro_CXCL2, and Macro_DNAJB1 (Fig. 4D). Heatmaps displayed the top five most highly upregulated genes in each cell type (Fig. 4E). Calculating Entosis scores for macrophage subtypes revealed the highest score in the Macro_CXCL2 subtype, followed by Macro_ISG15 (Fig. 4F). Moreover, compared to the low Entosis score group, the high Entosis score group showed the strongest association with the CXCL2 subtype (Fig. 4G). This suggests that the Macro_CXCL2 subtype may represent a core functional subtype characterized by high Entosis, and its strong correlation with Entosis may play a key role in regulating the tumor immune microenvironment. Based on this, the study employed the “FindAllMarkers” tool to identify marker genes for the Macro_CXCL2 subtype and conducted GO enrichment analysis. Results showed significant enrichment at the BP level in biological processes such as response to bacterial components, leukocyte migration, and chemotaxis; at the CC level in cellular components including secretory granules, outer plasma membrane, and transcription complexes; and at the MF level, they were significantly enriched in molecular functions including cytokine/chemokine activity, transcriptional regulation activity, and receptor binding (Fig. 4H). These findings indicate that the Macro_CXCL2 subtype represents a key cellular population mediating local immune inflammatory responses, suggesting it may serve as a core cellular bridge linking entosis-associated cellular stress to the anti-tumor immune microenvironment.

Pseudotime analysis

To further elucidate the origin and developmental differentiation patterns of macrophages and clarify the influence of entosis status on their differentiation process, this study employed the Monocle 2 algorithm to conduct pseudo-time trajectory analysis. Macrophages were sequenced based on gene expression characteristics and transition patterns (Fig. 5A). Results revealed that the high-Entosis-score macrophage subpopulation (EnhighMacro) exhibited significantly higher proportions of cells in states 7, 11, and 13, indicating that this subpopulation predominantly resides in late differentiation stages. Conversely, the low-Entosis-score subpopulation (EnlowMacro) showed higher proportions of cells in states 1, 2, and 4, corresponding to early differentiation stages (Fig. 5B). To validate this finding, we further quantified differences in the fitted time series of cell populations with varying Entosis scores across tissues. Violin plot analysis revealed significantly higher fitted time series values for cells in the high Entosis group compared to the low Entosis group (Fig. 5C). This further confirms that high Entosis activity drives macrophage differentiation toward later stages, providing direct evidence for Entosis's regulatory role in macrophage differentiation. Analysis of differentiation differences across seven macrophage subtypes (Fig. 5D) revealed through bar charts that Macro_CXCL2 subtype cells predominantly enriched states 11, 8, and 7 (Fig. 5E), indicating this subtype resides at the terminal end of macrophage differentiation. Pseudo-time sequence color maps of cells from different tissue origins revealed Branch 6 as a critical node in Macro_CXCL2 subtype differentiation. BEAM (Branch Expression Analysis Modeling) analysis at this node identified Entosis-related differentially expressed genes, with heatmap visualization showing their dynamic expression patterns (Fig. 5F). Notably, core Entosis-related genes (CDC42, CXCL8, EZR, LAMP1, PTK2, and RHOA) exhibited a progressive increase in expression levels as cellular differentiation advanced (Fig. 5G). These findings suggest that these core genes may play a pivotal role in the terminal differentiation and entosis functional regulation of the Macro_CXCL2 subtype, laying the foundation for elucidating the differentiation regulatory mechanisms of this subtype.

Fig. 5.

Pseudotime analysis of macrophages. (A) Pseudo-time analysis of macrophages. (B) Pseudo-time analysis of EnhighMacro and EnlowMacro. (C) Violin plot showing differences in pseudo-time sequences between high Entosis and low Entosis. (D) Pseudo-time analysis of seven macrophage subtypes. (E) Bar chart displaying distribution of pseudo-time states across seven macrophage subtypes. (F) Heatmap illustrating differential expression patterns of Entosis-associated genes during cellular development. (G) Differential expression dynamics of representative genes over pseudo-time.

Cell-cell communication analysis

We utilized the CellphoneDB tool in conjunction with scRNA-seq data to analyze interactions between macrophages and other clustered cells. Results revealed that compared to the low Entosis group, the high Entosis group exhibited significantly increased numbers and strengths of interactions between macrophages and other cellular units (Fig. 6A). Furthermore, we examined ligand-receptor interactions between macrophages and stromal cells (including fibroblasts, myofibroblasts, and endothelial cells), identifying a communication mechanism mediated by the SPP1-CD44 axis (Fig. 6B). The expression levels of the SPP1-CD44 axis associated with this interaction were lower in the low Entosis group but relatively higher in the high Entosis group (Fig. 6C), suggesting that the SPP1-CD44 axis may play a key role in osteosarcoma development.

Fig. 6.

Analysis of intercellular communication in the TME. (A) Bar chart showing the number of communications between cell types in high- and low-Entosis groups. (B) Bubble plot illustrating significant ligand-receptor interactions between macrophages and stromal cells. (C) Circos diagram depicting the APP signaling pathway network. (D) Bubble plot showing significant ligand-receptor interactions between macrophages and immune cells, malignant cells. (E) Circos plot depicting the MHC-II signaling pathway network. (F) Heatmap displaying significant ligand-receptor interactions across different cell populations. (G) Scatter plot of signal reception and emission intensities in cell subpopulations. (H) Bubble plot showing significant ligand-receptor interactions between macrophages in high- and low-Entosis groups and malignant cells/osteoblasts. (I, J) Bar charts showing communication levels between macrophages in high- and low-Entosis groups and other cell types.

Subsequently, we analyzed differences in macrophage communication with immune cells and malignant cells. We found significantly enhanced TNF, MIF, and MHC-II signaling between macrophages and immune cells (B cells, T cells, NK cells) as well as malignant cells in the high Entosis group, indicating active immune recognition and synergistic killing effects between macrophages and adaptive immune cells within this microenvironment (Fig. 6D, E). To systematically visualize communication patterns between macrophage groups, we compared the intensity and number of interactions between macrophages and other cells in high- and low-Entosis groups via heatmap analysis (Fig. 6F). Scatterplot analysis further revealed that macrophages, monocytes, and NK/T cells served as primary signal-receiving cells in the high-Entosis group, while mesenchymal stem cells (MSCs) and fibroblasts acted as core signal-emitting cells (Fig. 6G).

To further investigate macrophage functional heterogeneity and their regulatory roles in the TME, we compared communication differences between high- and low-Entosis group macrophages with malignant cells and osteoblasts. Results showed significantly enhanced COL1A1/COL1A2-integrin signaling from low-Entosis macrophages to osteoblasts; while high-Entosis group macrophages exhibited significantly upregulated TNF-mediated killing signals toward malignant cells, along with regulatory signaling pathways such as NAMPT and LGALS9 (Fig. 6H). Further ligand-receptor interaction analysis revealed that high-Entosis group macrophages exhibited significantly more ligand-receptor interactions with other cell types compared to the low-Entosis group (Fig. 6I, J).

Construction of weighted gene coexpression networks

WGCNA is a systems biology approach for analyzing expression patterns of multiple genes across different samples, forming clusters or modules comprising genes with similar expression patterns. Genes within the same module may share biological functions, enabling investigation of associations between these modules and sample characteristics (e.g., clinical features). This study performed WGCNA on bulk RNA-sequencing (bulk RNA-seq) data from osteosarcoma patients on bulk RNA-seq data from osteosarcoma. Single-sample gene set enrichment analysis (ssGSEA) was used to calculate Entosis scores from the TARGET-OS cohort as a sample characteristic. The top 5,000 most variable genes in the expression profiles were selected for WGCNA analysis. A soft threshold of β = 5 (scaleless R² = 0.9) was chosen to achieve a scaleless network (Fig. 7A, B). After excluding gray modules via merged dynamic tree cutting, 11 gene co-expression modules were identified (Fig. 7C). Among these, the green, pink, and green-yellow modules showed high correlation with Entosis scores (Fig. 7D). Genes from these three key modules were extracted for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) functional enrichment analysis. Results revealed significant KEGG pathway enrichment in influenza and viral infections, allograft rejection, phagosomes, cytokine-cytokine receptor interactions, rheumatoid arthritis, viral protein and cytokine/cytokine receptor interactions, osteoclast differentiation, B-cell receptor signaling, and lysosomes (Fig. 7E). GO functional enrichment primarily involved biological processes such as leukocyte activation and adhesion, myeloid cell differentiation, viral defense response, negative regulation of viral processes, bone remodeling, bone resorption, and osteoclast differentiation (Fig. 7F).

Fig. 7.

Construction of WGCNA and identification of key modules correlated with Entosis score. (A) Optimal soft threshold. (B) Scale-free network. (C) Hierarchical clustering dendrogram of gene similarity based on topological overlap. (D) Correlation analysis between module signature genes and clinical features. (E) KEGG pathway analysis of key modules. (F) GO enrichment analysis of key modules.

Construction of prognosis models

To construct a gene signature associated with osteosarcoma prognosis, we identified 37 candidate signature genes by intersecting differentially expressed genes in high- and low-Entosis macrophages, WGCNA key module genes, differentially expressed genes between high- and low-Entosis groups in the TARGET-OS cohort, and genes related to the immune microenvironment (Fig. 8A). Using random sampling, 4/5 of the TARGET-OS whole cohort (n=85) was allocated to the training set (n=51), with the TARGET whole cohort and GSE21257 dataset serving as the validation set. First, univariate Cox proportional hazards regression analysis identified 10 genes significantly associated with osteosarcoma prognosis. To mitigate overfitting risk and optimize gene selection for osteosarcoma prediction, Lasso regression analysis further screened 6 candidate genes from these 10 (Fig. 8B, C). Subsequently, stepwise multivariate Cox analysis identified CD163, FCGR2A, NPC2, CD209, SAMHD1, and IKZF1 as independent prognostic factors. The risk score calculation formula is: Riskscore = CD163 × (-0.019) + FCGR2A × (-0.01744) + NPC2 × (-0.24167) + CD209 × (-0.0716) + SAMHD1 × (-0.13607) + IKZF1 × (-0.1268) . This formula was applied to each sample in the training set, the TARGET cohort, and the GSE21257 external validation set. Samples were then categorized into high-risk and low-risk groups using the median risk score across all samples in each dataset as the cutoff value.

Fig. 8.

Construction of the prognostic model. (A) Venn diagram showing the intersection of candidate genes. (B) Coefficients and penalty parameter (λ) of the Lasso model. (C) Corresponding coefficients for genes in the six modules. (D) KM survival curves for high- and low-risk groups in the training set. (E) ROC analysis for the training cohort. (F) Scatter plot of risk score distribution in the training set. (G) Kaplan-Meier survival curves for high- and low-risk groups in the TARGET cohort. (H) ROC analysis for the TARGET cohort. (I) Scatter plot of risk score distribution in the TARGET cohort. (J) Kaplan-Meier survival curves for high- and low-risk groups in the GSE21257 dataset. (K) ROC analysis for the GSE21257 cohort. (L) Risk score distribution scatter plot for the GSE21257 dataset. (M) Heatmap of enrichment differences in immune function-related gene sets between high- and low-risk groups. (N) Heatmap of enrichment differences in cell death-related gene sets between high- and low-risk groups. * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001.

In the training set, the Kaplan-Meier curves for high- and low-risk groups demonstrated significant survival differences (log-rank P < 0.001), with markedly shorter survival times in the high-risk group. The AUC values at 1, 3, and 5 years reached 0.668, 0.749, and 0.725, respectively, indicating robust discriminatory capability (Fig. 8D-F). Within the TARGET cohort, survival differences between high- and low-risk groups remained significant (log-rank P < 0.001), with significantly shorter survival times in the high-risk group. The 1-year, 3-year, and 5-year AUC values were 0.712, 0.753, and 0.703, respectively, further validating the model's robustness (Fig. 8G-I). Finally, in the external GEO validation dataset GSE21257, Kaplan-Meier curve analysis confirmed significant survival differences between high- and low-risk groups, with longer survival in the low-risk group. The 1-year, 3-year, and 5-year AUC values were 0.781, 0.730, and 0.710, respectively (Fig. 8J-L). Combining validation results across all datasets, this risk model demonstrates stable and robust predictive performance for survival in osteosarcoma patients.

Further analysis using GSVA based on immune function-related gene sets and cell death-related gene sets detected biological process differences between high- and low-risk groups. Results showed significantly elevated scores for WNT-β-catenin target genes and inhibitory ligands in the high-risk group, while most other immune function gene sets were activated in the low-risk group (Fig. 8M). Among cell death-related gene sets, Anoikis and Methuosis-related scores were elevated in the high-risk group, while other differentially expressed cell death-related gene sets showed increased scores in the low-risk group (Fig. 8N).

Validation of model genes

To further investigate the biological functions and clinical significance of the six prognostic core genes, this study grouped patients based on median gene expression levels and plotted Kaplan-Meier survival curves for the model genes. Results showed that the Kaplan-Meier survival curves for the high-expression group versus the low-expression group differed significantly for five of these genes (P<0.05), and high expression of all five genes was associated with favorable prognosis in osteosarcoma patients (Fig. 9A). Further analysis of the correlation between model genes and entosis scores revealed a strong positive correlation between model gene expression levels and entosis scores (Fig. 9B). This indicates that the constructed risk model exhibits good internal consistency and is closely associated with entosis-related protective biological processes.

Fig. 9.

Validation of model genes (A) Kaplan-Meier survival curves for model genes. (B) Correlation analysis between model gene expression and entosis scores. (C) AUC values for model genes at 3 years in the training set. (D) Expression distribution of model genes in scRNA-seq data. (E) Molecular subtyping of osteosarcoma patients based on the model gene signature (divided into two molecular clusters) when k=2. (F) Kaplan-Meier prognostic analysis for the two distinct molecular clusters. (G) PCA of the two distinct molecular clusters. (H) Differential expression of model genes across the two distinct molecular clusters. (I) GSVA analysis of biological functions in the two distinct molecular clusters. * p<0.05; ** p<0.01; *** p<0.001; **** p<0.0001.

To evaluate the clinical utility of the model genes, ROC curves were plotted for all six model genes. Results showed that the AUC values for each model gene in standalone diagnosis exceeded 0.7, while the combined diagnostic AUC for all model genes reached 0.803, indicating excellent diagnostic performance (Fig. 9C). Subsequently, we analyzed the expression distribution patterns of model genes using scRNA-seq data. Results revealed significantly lower expression levels of model genes in malignant tumor cells compared to mesenchymal stem cells (MSCs), osteoblasts, and chondrocytes (Fig. 9D). To independently validate the biological significance of model genes, we performed unsupervised clustering analysis using all six model genes. The analysis revealed the most pronounced differences between subgroups when k=2, indicating patients could be effectively divided into two clusters (Fig. 9F). OS significantly differed between clusters (P=0.0032, Fig. 9E): Cluster 2 correlated with favorable prognosis, while Cluster 1 associated with poor prognosis. Principal component analysis (PCA) results (Fig. 9G) showed clear separation between the two subtypes, confirming the reliability of the clustering. Further feature analysis revealed that model genes were highly expressed in C2 subtype samples (Fig. 9H), accompanied by extensive activation of immune function-related pathways. Concurrently, WNT-β-catenin signaling activity and immunosuppressive ligand scores were significantly reduced (Fig. 9I).

Entosis regulates macrophage polarization and remodels the immune microenvironment in osteoscoma

To investigate the intrinsic association between osteosarcoma progression and macrophage polarization, macrophages were incubated with supernatant from osteosarcoma cells, and the polarization phenotype of macrophages was examined. The results showed that these macrophages exhibited an M1-polarized phenotype, characterized by elevated expression of M1 markers and reduced expression of M2 markers (Fig. 10A). Flow cytometric analysis further confirmed this conclusion (Fig. 10B). Furthermore, we collected conditioned media from osteosarcoma cells with high Entosis and low Entosis phenotypes to treat macrophages, and detected the expression of related markers by qPCR. Compared with the low Entosis group, conditioned medium from high Entosis osteosarcoma cells significantly upregulated CD163, FCGR2A, and CD209, and markedly downregulated NPC2 and SAMHD1 in macrophages (Fig. 10C). Among these, NPC2 is involved in regulating macrophage recruitment, and SAMHD1 is a key molecule in innate immune response and macrophage function. These expression changes suggest that high Entosis tumor cells can induce functional polarization of macrophages in a paracrine manner.

Fig. 10.

Entosis regulates macrophage polarization and remodels the immune microenvironment in osteoscoma. (A) The mRNA expression levels of macrophage polarization markers in macrophages stimulated with osteosarcoma cell supernatant were detected by qPCR. (B) The protein expression levels of CD86 and CD206 in macrophages from n macrophages stimulated with osteosarcoma cell supernatant were detected by flow cytometry. (C) Conditioned media were collected from osteosarcoma cells with high or low Entosis phenotypes and then used to treat macrophages, followed by qPCR analysis of macrophage marker gene expression. (D) The mRNA expression and (E) secretion of CXCL12 were examined in osteosarcoma cells with differential Entosis levels by qPCR and ELISA. (F) Protein expression levels of LC3, LAMP1 and E-cadherin were detected by Western blot, and Entosis index was quantified in adherent culture, suspension culture, and suspension culture with ROCK inhibitor (Y-27632) treatment groups. Comparisons between two groups were performed using two-tailed Student’s t-test. Comparisons among multiple groups were analyzed by one-way ANOVA followed by post hoc test. * p<0.05; ** p<0.01; *** p<0.001.

Based on bioinformatic analysis, the Macro_CXCL2 subtype displayed the highest Entosis score and was most strongly correlated with the high Entosis group (Fig. 10D). We further examined the expression and secretion of CXCL12 in high and low Entosis osteosarcoma cells, and found that CXCL12 expression was significantly increased in high Entosis cells (Fig. 10E). Meanwhile, bioinformatic analysis indicated that Entosis can be co-activated with the hypoxia pathway, TNFα-NFκB pathway, and MAPK signaling pathway, which jointly regulate stemness and invasive properties of malignant cells. Given the above pathway connections, we performed Western blot to detect the expression of LC3, LAMP1, and E-cadherin, and quantified the Entosis index. The results demonstrated that suspension culture significantly enhanced Entosis levels in osteosarcoma cells, whereas treatment with the ROCK inhibitor (Y-27632) markedly suppressed Entosis formation (Fig. 10F). In summary, high Entosis tumor cells significantly regulate macrophage polarization and functional phenotypes through paracrine signaling, and coordinate with key signaling pathways to participate in remodeling the immune microenvironment of osteosarcoma.

Discussion

As a new type of programmed death, Entosis’ role in tumors is tissue-specific and complex. This study found that the prognosis of cancer patients in the Entosis high score group was significantly better than that in the low score group, which is consistent with some research results in breast cancer and colorectal cancer [21,22], suggesting that Entosis may play a tumor suppressor role in cancer. Further GSEA analysis showed that the Entosis high-scoring group was significantly enriched in immune-related pathways such as antigen processing and presentation, chemokine signaling pathways, and MHC-II molecule-related gene sets. At the same time, the average enrichment level of 25 immune function-related genes was significantly increased. This indicates that Entosis may enhance the body's anti-tumor immune response by activating the tumor immune microenvironment, thereby improving patient prognosis. This is different from previous research findings in liver cancer—in liver cancer, Entosis has been mostly proven to promote tumor progression. This study suggests that Entosis may play a tumor suppressor role in osteosarcoma. This difference may be due to the heterogeneity of the microenvironment of different tumor types.

In osteosarcoma, cancer cells interact with immune cells and stromal cells to form an immunosuppressive TME, thereby enhancing the immune evasion ability of cancer cells. Intratumoral heterogeneity is also an important feature of osteosarcoma, leading to treatment resistance and differences in treatment outcomes between patients [23]. Understanding cancer cell heterogeneity and dynamic tumor immune microenvironment may provide new targets for osteosarcoma treatment. Single-cell sequencing analysis is a powerful tool for analyzing the cellular heterogeneity of the tumor microenvironment. This study found through single-cell data analysis that macrophages are the cell subpopulation most closely related to Entosis score, and their Entosis score is significantly higher than that of other cell types. Macrophages, as the core immune cells in the tumor microenvironment, have a high degree of functional heterogeneity and can be divided into the tumor-promoting M2 type and the tumor-suppressing M1 type [24]. This study further subdivided macrophages into 7 subtypes, among which the Macro_CXCL2 subtype had the highest Entosis score and was most strongly correlated with the high Entosis score group. GOGO enrichment analysis showed that the Macro_CXCL2 subtype was significantly enriched in functions related to leukocyte migration, chemotaxis, and cytokine/chemokine activity, suggesting that it may recruit effector immune cells (such as T cells and NK cells) to infiltrate tumor tissues through the secretion of chemokines and enhance anti-tumor immunity. Pseudo-time analysis further confirmed that high Entosis activity can promote the progression of macrophages to the late stage of differentiation, and the Macro_CXCL2 subtype is in the terminal stage of macrophage differentiation. The expression of core Entosis-related genes (such as CDC42, CXCL8) gradually increases with the differentiation process, suggesting that these genes may regulate the terminal differentiation and functional activation of the Macro_CXCL2 subtype. This discovery clarifies for the first time the functional subtypes of macrophages related to Entosis in osteosarcoma, providing a precise target for therapeutic strategies that target macrophages to regulate Entosis activity.

The stemness maintenance and differentiation potential of malignant cells are the core driving forces of tumor progression and metastasis [25]. This study found that the Entosis high score area highly overlaps with the CytoTRACE2 high score area, and the high Entosis score group accounts for a higher proportion of pluripotent cells. There is a significant positive correlation between the Entosis score and the CytoTRACE2 score, suggesting that Entosis may regulate the differentiation potential of malignant cells. Pathway analysis showed that malignant cells with high Entosis scores were significantly enriched in the hypoxia pathway, TNFα-NFκB and MAPK signaling pathways. These pathways have been confirmed to be closely related to the maintenance of tumor cell stemness, invasion and metastasis [26]. As a typical feature of the tumor microenvironment, hypoxia can enhance the stemness of cancer stem cells by activating the HIF-1α pathway [27,28]; while the TNFα-NFκB pathway promotes tumor cell proliferation and metastasis by regulating downstream target genes [29]. The coordinated activation of Entosis and these pathways may jointly regulate the stemness and invasion capabilities of malignant cells.

Intercellular communication analysis reveals the molecular mechanism by which Entosis regulates the osteosarcoma microenvironment. The number and intensity of interactions between macrophages and other cells were significantly increased in the high Entosis group. Among them, the communication between macrophages and stromal cells mediated by the SPP1-CD44 axis, and the communication between macrophages and immune cells/malignant cells mediated by TNF and MHC-II signals were significantly enhanced. As a secreted glycoprotein, SPP1 has been confirmed to promote the progression of various tumors by binding to CD44 [30]. In this study, the high activity of the SPP1-CD44 axis in the high Entosis group may be involved in the coordinated regulation of macrophages and stromal cells, affecting the homeostasis of the tumor microenvironment. As a classic inflammatory signaling pathway, TNF signaling can activate the killing function of effector immune cells, while MHC-II molecules participate in the antigen presentation process and enhance T cell recognition of tumor cells [31,32]. The enhancement of both in the high Entosis group further supports the mechanism of Entosis enhancing anti-tumor immunity by promoting communication between macrophages and immune cells. In addition, the TNF killing signal from macrophages in the high Entosis group to malignant cells was significantly upregulated, while the COL1A1/COL1A2-integrin signal from macrophages in the low Entosis group to osteoblasts was enhanced, suggesting that the Entosis state can regulate the functional bias of macrophages, thereby affecting tumor progression and bone microenvironment remodeling.

Building a prognostic model based on multi-omics data is an important means to achieve accurate prognostic assessment of tumors. This study screened out 37 candidate genes by integrating differentially expressed genes between high and low Entosis macrophages, WGCNA key module genes, differentially expressed genes between high and low Entosis score groups and immune microenvironment-related genes. Finally, a 6-gene (CD163, FCGR2A, NPC2, CD209, SAMHD1, IKZF1) prognostic model was constructed through single-factor Cox, Lasso and multi-factor stepwise Cox regression analysis. The model showed good prognostic prediction performance in the training set, validation set and external validation set. Further analysis found that high expression of the model gene was associated with a good prognosis of cancer patients and was highly positively correlated with the Entosis score, suggesting that the model can be used as an effective surrogate indicator for Entosis-related prognostic features. Unsupervised clustering based on model genes can divide patients into two molecular subtypes with significantly different prognosis. Among them, C2 subtype (high expression of model genes) is accompanied by widespread activation of immune function-related pathways, while WNT-β-catenin signaling Pathway activity and immunosuppressive ligand scores were significantly reduced, which further confirmed the mechanism by which Entosis-related features affect patient prognosis by regulating the immune microenvironment. It also provided a basis for molecular classification and personalized treatment of cancer. It is worth noting that CD163 and FCGR2A in the model are both macrophage-related markers, which echoes the previously discovered central role of macrophages in the regulation of Entosis and further verifies the consistency of the research results.

To further validate the regulatory role of Entosis in the immune microenvironment of osteosarcoma, in vitro functional experiments were conducted in this study. The results demonstrated that osteosarcoma cell supernatant induced an M1-polarized phenotype in macrophages, suggesting a pro-inflammatory immune response tendency within the TME. Notably, conditioned medium from osteosarcoma cells with a high Entosis phenotype significantly upregulated the expression of CD163, FCGR2A, and CD209, while downregulating NPC2 and SAMHD1 in macrophages, indicating that high-Entosis tumor cells can drive the conversion of macrophages toward a specific functional phenotype through paracrine signaling. In line with bioinformatics analyses, the Macro_CXCL2 subtype exhibited the highest Entosis score in the high-Entosis group and showed a strong correlation with Entosis activity, further supporting the central role of this subtype in Entosis-mediated immune regulation. Moreover, the expression and secretion of CXCL12 were significantly increased in high-Entosis cells, suggesting their potential involvement in microenvironment remodeling through immune cell recruitment. Pathway analyses revealed coordinated activation of Entosis with the hypoxia pathway, TNFα-NFκB pathway, and MAPK signaling pathway, which collectively regulate the stemness and invasive properties of malignant cells. Western blot analysis further confirmed that suspension culture markedly enhanced the expression of LC3, LAMP1, and E-cadherin in osteosarcoma cells and increased the Entosis index, whereas treatment with the ROCK inhibitor Y-27632 effectively suppressed this process. In summary, high-Entosis tumor cells regulate macrophage polarization via paracrine signaling and synergize with multiple key signaling pathways to participate in the remodeling of the immune microenvironment in osteosarcoma. These findings provide experimental evidence for Entosis as a potential therapeutic target and offer new insights into combined immunotherapeutic strategies for osteosarcoma.

Conclusions

This study systematically reveals the tumor-suppressive role and immune regulatory mechanism of Entosis in osteosarcoma through integrated multi-omics analysis and in vitro experiments. High Entosis activity is significantly correlated with favorable patient prognosis and an activated immune microenvironment, with macrophages, particularly the Macro_CXCL2 subtype, identified as the core target cells. Mechanistically, high-Entosis tumor cells regulate macrophage polarization via paracrine signaling and synergize with the hypoxia, TNFα-NFκB, and MAPK pathways to remodel the immune microenvironment. Furthermore, a six-gene prognostic model based on Entosis-related genes demonstrates robust predictive performance across multiple datasets, providing an effective tool for personalized prognosis assessment. This study offers a theoretical foundation and experimental evidence for Entosis-targeted combination immunotherapy strategies in osteosarcoma.

Ethics approval and consent to participate

The study was approved by the ethics committee of the Second Hospital of Hebei Medical University (2026-Y173). All analyses were performed using publicly available datasets (TCGA, GEO) and commercially obtained cell lines. No human subjects or animal experiments were directly involved in this study. The study was conducted in accordance with the Declaration of Helsinki.

Consent to publish declaration

Not applicable.

CRediT authorship contribution statement

Guoling Huang: Formal analysis, Software, Visualization, Writing – original draft. Yongbo Xiao: Data curation, Investigation, Methodology, Writing – original draft. Hongliang Zhang: Funding acquisition, Project administration, Supervision, Validation, Writing – review & editing. Yu Liu: Conceptualization, Funding acquisition, Project administration, Resources, Writing – review & editing.

Declaration of competing interest

The authors declare no potential competing interests.

Data availability

• The data that support the findings of this study are available from the corresponding author upon reasonable request.