We performed differential gene analysis between osteosarcoma and normal tissue samples using DEseq2 in R version 4.0.1 (Lucent Technologies Inc., Union, NJ, USA, https://www.r-project.org/). The samples were labeled as “normal” and “tumor.” We calculated the fold change (FC) for genes and subsequently filtered the differentially expressed genes (DEGs) based on the following criteria: adjusted p < 0.05, log₂FC > 1, or log₂FC < −1. Subsequently, we used the ggplot package (https://cran.rproject.org/web/packages/ggplot2/index.html) and the ‘pheatmap’ package (https://cran.r-project.org/web/packages/pheatmap/index.html) in R to generate volcano plots and heatmaps. Additionally, we performed Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for the DEGs with the ‘clusterProfiler’ package in R version 4.0.1, with a significance threshold set at a p-value below 0.05 (Lei et al., 2015).

This study extracted osteosarcoma gene microarray data, including gene expression profiles from six osteosarcoma samples and six normal tissue samples, from the GEO database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126209) at the National Center for Biotechnology Information (NCBI) in the United States.

We utilized the WGCNA analysis package in R version 4.0.1, with the aim of identifying gene modules that are highly correlated with osteosarcoma (Horvath and Dong, 2008). We computed scale independence (R2) and average connectivity to determine the soft threshold that conforms to an unscaled distribution, which was used for feature gene selection. Following this, we utilized hierarchical clustering and dynamic tree-cutting methods to discern gene modules, employing a substantial merging threshold of 0.25 to amalgamate akin gene modules. Ultimately, we computed both gene significance (GS) and module significance (MS) to evaluate the correlation and significance of genes in relation to clinical information.

We used the STRING website (https://string-db.org/) to perform PPI analysis on all genes within the green-yellow module. Subsequently, we conducted further analysis and visualization using the Molecular Complex Detection (MCODE) plugin in the Cytoscape software (http://cytoscape.org/).

In this study, we utilized the ‘glmnet’ package in R for LASSO regression analysis to screen core genes within the ‘greenyellow’ module (Park et al., 2015). LASSO regression analysis identifies variables by detecting the optimal λ value that minimizes classification error. The optimal value of the penalty parameter λ was selected using tenfold cross-validation. Receiver Operator Characteristic (ROC) curves were generated for predictive models via ‘timeROC’ package in R. The Area Under the Curve (AUC) was computed to gauge the predictive efficacy of the prognostic gene model. All statistical analyses were carried out through R software, with significance determined at p < 0.05 for all tests. Optimal cutoff points for risk scores and survival time were determined using the ‘maxstat’ package in R, and the training dataset was stratified into high-risk and low-risk groups based on these cutoff points. We utilized Kaplan-Meier analysis to compare overall survival times between the high-risk and low-risk groups. Survival curves for these groups were constructed with the ‘survminer’ and ‘survival’ packages in R, and inter-group differences were evaluated through chi-square testing.

The cell lines used in this study included osteosarcoma cells (MG63 cells, catalog number CL-0157; U2OS cells, catalog number CL-0236; HOS cells, catalog number CL-0360) and human fetal osteoblast hFOB1.19 cells (CL-0353), all of which were purchased from Procell Technology Co., Ltd. (China). These cells were centrifuged during cultivation and then supplemented with 1 mL of Dulbecco’s Modified Eagle Medium (DMEM) medium (C11995500BT, GIBCO, Thermo Fisher Scientific, Inc., USA) containing 10% fetal bovine serum (FBS) (Peak Inc., USA), 1% penicillin-streptomycin (Beyotime Institute of Biotechnology, China), and 20 ng/mL epidermal growth factor (EGF) (catalog number P00033, Solarbio Science & Technology Co., Ltd., China). All cells were cultured at 37°C in a 5% CO₂ incubator.

According to the reagent kit instructions, TRIzol reagent (Thermo Fisher Scientific, Inc.) was used to extract total RNA (Wang et al., 2020). Subsequently, we employed Tsingke Biotechnology Co., Ltd.'s Goldenstar™ RT6 cDNA Synthesis Kit Ver.2 (TSK302M, China) for reverse transcription, with the reaction conditions set at 25°C for 10 min, 55°C for 30 min, and 85°C for 5 min. To assess mRNA expression levels, we used the 2×T5 Fast RT-qPCR Mix (SYBR Green I) reagent kit from Tsingke Biotechnology Co., Ltd. (TSE002, China), with reaction conditions of 95°C for 30 s, followed by 40 cycles of 95°C for 3 s and 60°C for 30 s. We selected human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as the reference gene and calculated relative gene expression levels using the 2^−ΔΔCt method. Specific primer information for RT-qPCR is presented in Table 1.

In fixed cell cultures, 0.3% Triton X-100 (ST795, Beyotime Institute of Biotechnology, China) was added and incubated at 37°C for 5 min. Goat serum (C0265, Beyotime Institute of Biotechnology, China) was then added and incubated at room temperature for 60 min. Subsequently, the sections were incubated with antibodies against β-catenin (A19657, ABclonal Technology, China), CHMP4C (bs-7744R, Bioss Biotechnology Co., Ltd., China), GSK3β (A2081, ABclonal Technology, China), and p-GSK3β (bs-3161R, Bioss Biotechnology Co., Ltd., China) overnight at 4°C. After washing with Phosphate-Buffered Saline (PBS) (G0002, Sevier Biotechnology Co., Ltd., China), FITC Goat Anti-Rabbit IgG (H + L) (AS024, ABclonal Technology, China) was applied and incubated in the dark at 25°C for 1.5 h. Sections were subjected to 4′,6-diamidino-2-phenylindole (DAPI) staining (C1005, Beyotime Institute of Biotechnology, China), followed by a 5-min incubation in the dark and removal of excess DAPI through PBS wash. Subsequently, these sections were sealed with an anti-fade mounting medium (P0126; Beyotime Institute of Biotechnology, China). Ultimately, an inverted fluorescence microscope (ICX41; Sunny Optical Technology (Group) Co., Ltd., China) was employed to observe and capture images of the sections.

The lentiviral overexpression vector pLVX-IRES-puro-CHMP4C and interference vector pLVX-shRNA1-CHMP4C were constructed by Chongqing Biomedicine Co., Ltd. (China). MG63 cells were passaged at a density of 1×10⁵ cells/well in a 24-well plate. After 24 h of incubation with fresh culture medium, the control group received 30 μL of empty viral suspension and 1 μL of polybrene (ZY140621, ZeYe Biotechnology Co., Ltd., China), whereas the experimental group received 30 μL of either overexpression lentivirus or interference lentivirus suspension and 1 μL of polybrene. The cells were then placed in a cell culture incubator for 48 h. Subsequently, the cells were continuously passaged, and the culture medium was changed every 2 days and supplemented with 5 μg/mL puromycin (ST551-10, Beyotime Institute of Biotechnology, China) for selection to obtain CHMP4C overexpressing MG63 cells (OE-CHMP4C), CHMP4C interference MG63 cells (IN-CHMP4C), and their corresponding empty vector control MG63 cells (OE-NC, IN-NC).

Cells in the logarithmic growth phase were harvested and plated into 96-well plates. Following that, 10 μL of CCK-8 solution (C0038; Beyotime Institute of Biotechnology, China) was introduced to each well. After a 1-h incubation at 37°C in a cell culture incubator, the absorbance of each well was assessed at 450 nm with a microplate reader (CMax Plus, Molecular Devices Instruments Co., Ltd., USA).

Cell viability (100%) was calculated as follows:

Experimental group: Absorbance of cells overexpressing or silencing plasmids along with CCK-8 solution.

Blank group: Absorbance of wells containing culture medium and CCK-8 solution without cells.

Control group: Absorbance of wells containing empty vector-transfected cells and CCK-8 solution. Cell viability  100 % = experimental group – blank group / control group – blank group × 100

Logarithmically growing CHMP4C-overexpressing MG63 cells, CHMP4C-knockdown MG63 cells, and their respective empty vector control cells were harvested to prepare the cell suspensions. The cell suspensions were serially diluted and seeded into six-well plates containing the culture medium, followed by a 6-day incubation period. Upon the emergence of macroscopic colonies in the culture dishes, the cultivation process was concluded. The supernatant was removed, and the cells were fixed with 4% paraformaldehyde (DF0135, Leagene Biotechnology Co., Ltd., China) for 20 min. Subsequently, staining with crystal violet staining solution (DZ0053, Leagene Biotechnology Co., Ltd., China) was carried out for 15 min. Ultimately, the clone formation rates were computed. Clone efficiency = number of clones  /  number of seeded cells × 100

CHMP4C-overexpressing MG63 cells, CHMP4C-silenced MG63 cells, and their respective empty vector control cells were adjusted to a density of 5×10⁵/mL cells/mL in serum-free medium and seeded in Transwell chambers (3,421, Corning Inc., USA). The lower chambers were loaded with DMEM medium containing 10% FBS and 1% streptomycin, and then placed in a cell culture incubator for 24 h. Following incubation, Transwell chambers were extracted, and the culture medium was aspirated. The cells were washed twice with calcium-free PBS, fixed with 4% paraformaldehyde for 20 min, and stained with 0.1% crystal violet for another 20 min. The upper layer of the non-migrated cells was gently wiped off with a wet cotton ball. After washing with PBS, cells from three randomly selected fields were counted, and the average was calculated under a 100x microscope (CKX3-SLP, OLYMPUS, Japan).

Thirty-two male specific pathogen free (SPF)-grade BALB/C nude mice, approximately 6 weeks old, were obtained from Chongqing Ensiweier Biological Technology Co., Ltd., China. Mice were randomly divided into four groups (OE-NC, OE-CHMP4C, IN-NC, and IN-CHMP4C), each consisting of eight mice. The mice were acclimated in an animal facility at 25°C and a 12–12 h light-dark cycle for 1 week. MG63 cells overexpressing CHMP4C, MG63 cells with CHMP4C knockdown, and their corresponding control MG63 cells were digested with trypsin (S310KJ, Beyotime Institute of Biotechnology, China) to prepare cell suspensions at a cell density of 3×10⁷ cells/mL. The skin of the mice at the thigh root was disinfected and each mouse was subcutaneously injected with a cell suspension (0.5 mL). After 4 weeks, the mice were anesthetized and tumor images were observed and captured. Tumor dimensions were gauged, and euthanasia of the mice was carried out. The tumors were fully excised and weighed. Following that, RT-qPCR and immunofluorescence techniques were employed to assess the mRNA and protein expression levels. The animal study protocol was approved by the Ethics Committee of Affiliated Hospital of Qingdao University (QYFY WZLL 28188, 2023.11.06).

GraphPad Prism 8.0 (GraphPad Company, San Diego, CA, USA) was used for data analysis. The results are presented as mean ± standard deviation. Each experiment was performed at least three times. Independent sample t-tests were used for comparisons between two groups, and one-way analysis of variance (ANOVA) and Tukey’s post hoc test were used for comparisons among three or more groups. A p-value <0.05 was considered statistically significant.

A total of 7,189 DEGs were identified, with 1,511 DEGs upregulated in normal tissue samples and 5,678 DEGs downregulated. Figure 1A presents the heatmap generated from hierarchical clustering analysis. The volcano plot illustrates the differential gene expression between osteosarcoma tissue and normal tissue (Figure 1B). To unveil the potential biological functions of DEGs, we conducted GO and KEGG analyses. Those enriched biological processes in the GO analysis included embryonic organ development, pattern specification process, and axon development. Enriched cellular components comprised the collagen-containing extracellular matrix, actin cytoskeleton, and cell-cell junction. Additionally, enriched molecular functions encompassed DNA-binding transcriptional activator activity, RNA polymerase II-specific, DNA-binding transcription activator activity, receptor ligand activity, and signaling receptor activator activity (Figure 1C). KEGG analysis revealed enrichment in pathways such as axon guidance, cytokine-cytokine receptor interaction, Wnt signaling, Rap1 signaling, and Mitogen-Activated Protein Kinase (MAPK) signaling (Figure 1D).

Through scale independence and average connectivity comparisons, we found that when the soft threshold was set to 17 for gene-to-gene connections (Figures 2A, B), we were able to construct a hierarchical clustering tree containing 10 key modules (Figure 2C). However, it is worth noting that genes within the gray module appeared to lack distinct functional similarities. To ascertain the significance of each module, we conducted correlation analysis between the obtained modules and clinical features. The results revealed a significant positive correlation between the green-yellow module and osteosarcoma (R = 0.82, p = 0.001) (Figure 2D), which encompassed 754 DEGs. To gain a deeper understanding of the biological functions of DEGs within the green–yellow module, we performed GO and KEGG enrichment analyses. GO functional enrichment encompasses biological processes, such as regulation of the response to DNA damage stimulus, regionalization, and positive regulation of protein localization. Concerning cellular components, enrichment was noted in the nuclear membrane, nuclear envelope, and cell leading edge. Molecular functions exhibited enrichment in DNA-binding transcriptional repressor activity, RNA polymerase II-specific DNA-binding transcription activator activity, DNA-binding transcription repressor activity, DNA-binding transcription activator activity, and RNA polymerase II-specific (Figure 3A). Furthermore, in the KEGG enrichment analysis, we identified associations with pathways such as axon guidance, Hippo signaling pathway, hepatocellular carcinoma, Wnt signaling pathway, and MAPK signaling pathway (Figure 3B).

We constructed and analyzed a protein-protein interaction network for the DEGs in the green–yellow module. We imported the DEG information into the STRING database and further analyzed the generated gene interaction network using Cytoscape software (Figure 3C). Utilizing the Maximal Clique Centrality (MCC) algorithm in the “Cytohubba” plugin, we identified the top 10 genes that exhibited the closest interactions within the gene network. These genes were Nucleolar Protein 58 (NOP58), WD Repeat Domain 43 (WDR43), Ribosomal Protein R9 (RPR9), WD Repeat Domain 12 (WDR12), WD Repeat Domain 36 (WDR36), U3 small nucleolar RNA-associated protein 18 (UTP18), Partner of NOB1 homolog (PNO1), Dyskerin Pseudouridine Synthase 1 (DKC1), Nucleolar and Coiled-body Phosphoprotein 1 (NOLC1), and Ribosomal RNA Processing 1 (RRS1) (Figure 3D). The darker color signifies a more significant role of these genes in the occurrence and development of osteosarcoma.

The optimal λ that produces the smallest classification error was identified in the LASSO model through 10-fold cross-validation. Subsequently, 15 hub genes with non-zero coefficients were identified (ArfGAP with dual PH domains 2, ADAP2; Carbonic Anhydrase III-Antisense RNA 1, CA3-AS1; Charged Multivesicular Body Protein 4C, CHMP4C; Family with sequence similarity 222 member B, FAM222B; FK506 Binding Protein 11, FKBP11; HOXA11 antisense RNA, HOXA11. AS; PiggyBac transposable element derived 5, PGBD5; Phorbol-12-Myristate-13-Acetate-Induced Protein 1, PMAIP1; Proline and serine rich 2, PROSER2; Ring Finger Protein 38, RNF38; SH3 Domain Binding Protein 2, SH3BP2; Small Nucleolar RNA-H/ACA Box 12, SNORA12; tRNA Methyltransferase 1 Homolog, TRMT1; Thioredoxin Like 4B, TXNL4B; Zinc Finger Protein 200, ZNF200) (as shown in Figures 4A, B). To enhance the scrutiny of the risk model’s accuracy in prognosis analysis, we generated ROC curves for the predictive model. The calculated AUC was 0.947, indicating a substantial predictive capability (Figure 4C). This suggests that the risk prognosis model has a high predictive accuracy. To affirm the association between the expression of these 15 genes and OS patients, Kaplan-Meier survival analysis was employed. Patients were stratified into high- and low-expression groups, determined by the median gene expression values. Our outcomes revealed statistically significant differences in survival rates between low- and high-expression groups for the CHMP4C, FAM222B, PROSER2, SNORA12, and ZNF200 genes (p < 0.05) (Figure 4D). Notably, CHMP4C displayed the most substantial difference (p = 0.0094), signifying a robust correlation between CHMP4C expression and the prognosis of OS patients. Individuals in the low-expression group of CHMP4C exhibited a more favorable prognosis with higher survival rates (p < 0.05). The survival curves for the other nine genes are shown in Figure 5.

In this study, we conducted an analysis of CHMP4C and the related proteins GSK3β, p-GSK3β, and β-catenin. CHMP4C is a protein associated with the endocytic pathway and closely linked to processes such as cell division, endocytosis, and signal transduction (Zhang et al., 2022). Research has indicated a significant association between the GSK3β/β-catenin signaling pathway and osteosarcoma proliferation (Ruan and Zhao, 2018). In the RT-qPCR analysis, we observed distinct expression patterns of CHMP4C and β-catenin across the four OS cell lines. Specifically, CHMP4C mRNA expressions were markedly elevated in U2OS, HOS, and MG63 cells compared to hFOB1.19 cells (p < 0.01), with MG63 cells exhibiting the highest expression among them. Conversely, β-catenin mRNA expression levels increased sequentially in hFOB1.19, U2OS, HOS, and MG63 cells, with statistical significance compared to hFOB1.19 (p < 0.01) (Figure 6A). Our outcomes of immunofluorescence staining revealed notable disparities in the protein expression levels of CHMP4C, p-GSK3β, and β-catenin among the U2OS, HOS, and MG63 osteosarcoma cells, with significantly higher expression observed in these cells compared to hFOB1.19 cells (Figures 6B−D).

In our exploration of the biological function of CHMP4C in OS cells, MG63 cells were transfected with a synthetic CHMP4C over-expression vector, CHMP4C interference lentiviral vector, and an empty vector for validation. RT-qPCR results revealed that, in comparison to empty vector group, CHMP4C over-expression notably augmented the mRNA expressions of CHMP4C and β-catenin in MG63 cells (p < 0.0001) and concurrently diminished the expressions of GSK3β (p < 0.001). Conversely, CHMP4C knockdown significantly repressed the expression of CHMP4C and β-catenin in MG63 cells (p < 0.001) while elevating GSK3β expression (p < 0.0001) (Figure 7A). Immunofluorescence staining further validated these findings. CHMP4C over-expression notably heightened the protein expression of CHMP4C (Figure 7B), p-GSK3β (Figure 7D), and β-catenin (Figure 7E) in MG63 cells, while concurrently reducing GSK3β expression (Figure 7C). In contrast, CHMP4C knockdown greatly curtailed the protein expression of CHMP4C (Figure 7B), p-GSK3β (Figure 7D), and β-catenin (Figure 7E) in MG63 cells, while elevating GSK3β expression (Figure 7C).

Our CCK-8 assay outcomes demonstrated that, in comparison to the control group, CHMP4C over-expression significantly heightened the proliferation of MG63 cells, whereas interference with CHMP4C expression markedly curtailed MG63 cell proliferation (p < 0.0001) (Figure 8A). Findings from the colony formation assay revealed that the number of MG63 cells and cloning efficiency were notably elevated in the CHMP4C over-expression group in contrast to control group (p < 0.001) (Figure 8B). Conversely, the number of MG63 cells and cloning efficiency were substantially reduced in CHMP4C interference group as compared to control group. In Transwell migration assays, compared to control group, CHMP4C over-expression resulted in a significant augmentation in the number of migrating MG63 cells (p < 0.0001), whereas interference with CHMP4C expression led to a notable reduction in the number of migrating MG63 cells (p < 0.001) (Figure 8C).

We further investigated the tumor occurrence through subcutaneous tumor experiments in nude mice. Compared to the empty vector group, the CHMP4C overexpression group showed a significant increase in tumor weight and volume (p < 0.0001), while the CHMP4C knockdown group exhibited a significant decrease in tumor weight (p < 0.001), with no significant difference in tumor volume (p > 0.05) (Figures 9A, B). RT-qPCR analysis of mRNA expression in tumor tissues revealed that compared to the empty vector group, CHMP4C overexpression significantly increased the expression of CHMP4C and β-catenin in mouse tumor tissues (p < 0.0001) and significantly decreased GSK3β expression (p < 0.001). Conversely, interference with CHMP4C expression significantly reduced the expression of CHMP4C and β-catenin, and increased GSK3β expression (p < 0.0001) (Figure 9C). Our outcomes of immunofluorescence staining revealed that the protein expressions of CHMP4C, p-GSK3β, and β-catenin were notably elevated in CHMP4C over-expression group compared to empty vector group. Conversely, the GSK3β protein expression was significantly lower in CHMP4C over-expression group. In contrast, in CHMP4C knockdown group, the protein expressions of CHMP4C, p-GSK3β, and β-catenin were markedly lower than those in empty vector group, while the GSK3β protein expression was significantly higher (Figure 9D).