Constructing anoikis-related signatures in glioma and identifying PLAU as a new prognostic biomarker

Abstract

Background

Glioma, the most common aggressive tumor found in the central nervous system, is linked to a poor prognosis. Anoikis is a form of cell death that arises from the detachment of cells. Disruption of the anoikis pathway can facilitate the survival and dispersal of cancer cells from the primary tumor site, thus playing a critical role in the spread and advancement of cancer. The process of Epithelial-to-mesenchymal transition (EMT) is crucial in the progression and spread of cancer, where epithelial cells undergo a loss of polarity and adhesion, transitioning into mesenchymal cells with heightened capability for invasion and metastasis. However, there is a dearth of research examining the mechanism of anoikis and its correlation with the EMT process in glioma.

Methods

In our study, data was extracted from 1317 samples of glioma for comparative examination. Through the utilization of 28 genes associated with resistance to cell detachment, we aimed to distinguish various categories of glioma patients by evaluating the immune environment and pathways of enrichment within the two subgroups. By implementing diverse statistical methodologies such as COX regression analysis and the least absolute shrinkage and selection operator regression, we developed a risk scoring system to categorize glioma patients into cohorts of high and low risk. The clinical features, infiltration of immune cells, and responsiveness to drugs were thoroughly investigated for both categories, highlighting distinctions across various domains. Moreover, our investigation encompassed in vitro trials to assess the function and expression of the pivotal gene, PLAU.

Results

The differences in prognoses, immune status, and drug sensitivities between high-risk and low-risk groups highlight the need for personalized treatment approaches in glioma patients. By integrating risk scores with clinicopathological characteristics, a nomogram was created to better predict patient outcomes and guide treatment decisions. The nomogram demonstrated its usefulness through decision curve analysis, showing potential benefits for enhancing clinical strategies in glioma management. In cell experiments, targeting Plasminogen activator urokinase (PLAU)led to a notable decrease in the growth, spread, and movement abilities of T98G and U251 cell lines, as well as blocking the epithelial-mesenchymal transition process.

Conclusion

The study we conducted establishes the groundwork for comprehending the function of anoikis genes in glioma, and recognizes PLAU as a potential biomarker for glioma.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12920-025-02288-0.

Introduction

Glioma, the most prevalent tumor within the central nervous system (CNS), originates from the layer of neuroepithelial cells. In the year 2021, the fifth edition of the World Health Organization Classification of Tumors originating in the Central Nervous System categorized gliomas based on their natural history [1]. Low-grade gliomas, also known as LGG, are divided into grades 1 and 2, while high-grade gliomas are categorized as grades 3 and 4. Grade 2 encompasses oligodendrogliomas and astrocytomas. Grade 3 includes anaplastic oligodendroglioma, anaplastic astrocytoma, pleomorphic xanthoastrocytoma, and anaplastic ependymoma. Glioblastoma (GBM), the most malignant form of glioma, falls under grade 4 [2]. The primary treatment modalities for patients with glioma include surgical resection, radiotherapy, and chemotherapy [3]. The aggressive invasive characteristics of glioma cells, leading to widespread infiltration into neighboring brain tissue, contribute significantly to the grim prognosis and high mortality rates associated with gliomas [4]. Nonetheless, the intricate molecular mechanisms governing the invasion and migration of glioma cells remain not fully elucidated [5, 6]. Despite some advancements in glioma research, there has been minimal progress in enhancing glioma prognosis, with no significant breakthroughs recorded thus far.

Anoikis is a programmed cell death mechanism triggered by the loss of cell adhesion, specifically due to the loss of attachment between cells themselves or between cells and the extracellular matrix (ECM). Its primary role is to prevent abnormal cell proliferation of exfoliated cells and maintain tissue balance [7]. Nevertheless, when tumor cells develop resistance to Anoikis, they can evade programmed cell death, leading to a critical role in the invasion and spread of tumors [8]. Recent research has identified a link between Anoikis-related genes and the outlook for tumors. For example, MYH9 stimulates CTNNB1 transcription, which results in resistance to Anoikis in gastric cancer [9]. Similarly, HRC hinders endoplasmic reticulum stress in hepatocellular carcinoma, promoting resistance to Anoikis and metastasis [10]. Furthermore, the Nm23-ITGA5 pathway has been implicated in breast cancer cell invasion, indicating that regulation of this pathway could potentially impede the establishment of metastatic breast cancer cells [11]. Despite these important discoveries, there is a dearth of comprehensive studies that have assessed the impact of Anoikis on glioma. Hence, it is crucial to explore the cellular characteristics and molecular mechanisms of Anoikis in glioma, as this could significantly enhance the development and prognosis of glioma treatment.

EMT is a fundamental biological event where malignant tumor cells, originating from epithelial cells, acquire both migratory and invasive properties [12]. This transformation involves the loss of cell polarity and adherence in epithelial cells, causing them to transition into mesenchymal cells. The disruption of cell-to-cell connections in mesenchymal cells enables them to take on a phenotype similar to that of mesenchymal stem cells, thus increasing their invasive and metastatic potential [13]. Glioma cell EMT is distinguished by the downregulation of epithelial markers such as E-cadherin, as well as the upregulation of mesenchymal markers like N-cadherin and Vimentin [14]. Various elements present within tumor cells control the EMT process [15]. Upon being triggered by these factors, a chain reaction is initiated, enhancing cell migration and invasion abilities [16, 17]. Consequently, blocking the EMT process is crucial for effective glioma treatment. Nevertheless, the precise impact of Anoikis-related genes on glioma's EMT process is not yet fully understood.

In this investigation, a variety of bioinformatics techniques were employed to analyze sequencing data sets derived from glioma patient tissues. The primary objective was to pinpoint genes that are differentially expressed in relation to anoikis and explore its presence in various subtypes of glioma. A prognostic scoring system was developed based on the link between genes associated with apoptosis and both patient outcomes and the tumor immune microenvironment (TME). The validity and reliability of this risk score model were confirmed. Additionally, an examination was conducted on the connection between ANRGs, the immune microenvironment, patient prognosis, and responsiveness to chemotherapy. Through the examination of genetic data, the significance of ANRGs in evaluating the prognosis of glioma patients was revealed. This discovery has profound implications for enhancing treatment approaches for glioma.

Materials and methods

Gene expression and clinical data acquisition

RNA-seq data from The Cancer Genome Atlas (TCGA) database were utilized to analyze glioblastoma (GBM) and low-grade glioma (LGG). In total, 705 glioma tissue samples were examined, along with 432 normal control samples sourced from the Genotype-Tissue Expression (GTEx) website. Moreover, mRNA expression profiles from glioma patients (23 non-tumor and 157 tumor) were obtained from the Gene Expression Omnibus (GEO) database. The gene expression data underwent batch correction and integration through the use of the “limma” and “sva” packages [17]. Clinical information on 1158 glioma patients (1108 cases from TCGA and 50 cases from GSE43378) was retrieved from the TCGA and GEO databases.

Identification of differentially expressed genes

A total of 516 genes related to anoikis, referred to as ANRGs, were gathered from the GeneCard database (https://www.genecards.org/) and the Harmonizome portal (https://maayanlab.cloud/Harmonizome/). Moreover, the TCGA-GTEx and GSE4290 datasets were used for differential expression analysis of genes that are differentially expressed (DEGs) with the R software (version 4.1.3) [18] utilizing the 'limma' package [19]. Criteria for analysis included a threshold of |log2FC|> 1.0 and FDR < 0.05.

Consensus clustering

The utilization of consensus clustering aimed to identify unique correlation patterns of anoikis associated with expressions from the k-means method. To validate the clustering reliability, the uniform manifold approximation and projection (UMAP) were employed with the R package 'ggplot2'.

Functional enrichment analysis

In order to carry out GSVA analysis, we acquired the file titled 'c2.cp.kegg.v7.4.symbols.gmt'. The analysis for GSVA enrichment was executed utilizing the 'GSVA' package in R [20]. Furthermore, we conducted functional enrichment analysis utilizing the 'clusterProfiler' package within the R software (version 4.1.3) (https://www.r-project.org). This assessment encompassed an analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) as well as Gene Ontology (GO).

Development and validation of prognostic signatures based on anoikis-related genes

Genes correlated with survival were identified through the use of Univariate Cox regression analysis. Variables showing significance with p values < 0.01 were chosen for further investigation employing the least absolute shrinkage and selection operator (LASSO) regression technique. The 'glmnet' software package in R was applied to minimize the number of genes in the final risk assessment model. Following this, the genes derived from the LASSO regression were integrated into the multivariate Cox regression analysis to establish the risk score formula: risk score = ∑ (ð × Exp), where ð signifies the pertinent regression coefficient and Exp represents the mRNA expression value. By categorizing all patients based on the median risk score, they were divided into two groups: high-risk and low-risk. The performance of the model was assessed using Kaplan–Meier (KM) survival curve and time-dependent receiver operating characteristic (ROC) curve analysis. In order to corroborate the prognostic model, data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) database (GSE43378) were employed.

Relationship between risk score and immune cell infiltration

To quantify the tumor microenvironment (TME), we implemented Single-sample Gene Set Enrichment Analysis (ssGSEA) along with CIBERSORT [21]. The latter was used to estimate immune cell type proportions across both low-risk and high-risk cohorts. We normalized the total scores of the estimated immune cell types in each sample to a value of 1. Furthermore, to explore the relationship between risk scores and immune infiltrating cells, we performed a Spearman rank correlation analysis.

Construction and evaluation of predictive nomogram and drug susceptibility analysis

Nomograms were created by incorporating risk ratings and clinicopathological characteristics. Validation took place internally to confirm precision via calibration plots. Evaluating the accuracy of the nomogram involved employing the Time-C index. The assessment of clinical net benefit was carried out using decision curve analysis (DCA) [22]. Furthermore, the 'oncopredict' R package was utilized to explore the potential effectiveness of clinical medications in high-risk and low-risk cohorts, and to pinpoint chemotherapeutic agents suitable for glioma therapy [23].

Tumor immunity single cell hub database

The Single Cell Tumor Immunity Hub (SCTIH; http://sctih.comp-genomics.org) is a notable online repository that concentrates on single-cell RNA sequencing of the tumor microenvironment [24]. Researchers used this repository to carry out an in-depth analysis of the diversity within the tumor microenvironment across different databases and cell categories.

Cell culture and transfection

Cell lines of glioma (U87, T98G, A172, and U251) and human astrocytes (HA) were procured from the cell repository of the Chinese Academy of Sciences. Glioma cells were nurtured in DMEM (Biological, Salem, US) complemented with 10% Gibco FBS (GIBCO, US) and upheld at 37 °C with 5% CO2. Han Biotechnology (Shanghai, China) provided the siRNA directing PLAU, which was introduced into the cells via Lipofectamine 3000 (L3000015, Invitrogen). The PLAU siRNA sequences were 5′-GCUCUGAAGUCACCACCAATT-3′ and 5′-UUGGUGGUGACUUCAGAGCTT-3′. Subsequent to 48–72 h, the manipulated cells were harvested for further evaluation.

RT-PCR

The Superbrilliant assay (Zhongshi, Tianjin, China) was used to extract RNA from HA and U87, T98G, A172, and U251 cell lines. Following that, reverse transcription was carried out with the Takara kit (Bori Medical, Beijing, China), and PCR analysis was performed using the SYBR qPCR Mix kit (Vazyme, Nanjing, China). Supplementary Table 1 contains a list of all primer sequences utilized in this investigation, with every trial replicated thrice independently.

Western blot

Protein extraction involved the use of the protein extraction reagent (APExBIO, K1015). The concentration of the protein obtained was determined through BCA (bicinchoninic acid) assay. Following this, the samples were subjected to denaturing polyacrylamide gel electrophoresis, and the resulting protein bands were then transferred onto polyvinylidene fluoride membranes. The polyvinylidene fluoride membrane underwent a 15-min blocking step using a blocking solution. After removal of the blocking solution, the membrane was left to incubate overnight at 4 °C with primary antibodies. The primary antibodies employed were: PLAU (Abways Biotechnology, Shanghai, China, P00749), E-Cadherin (ABclonal Technology, Wuhan, China, A3044), N-Cadherin (Abways Biotechnology, Shanghai, China, P19022), Vimentin (Abways Biotechnology, Shanghai, China, P08670), and GAPDH (Abways Biotechnology, Shanghai, China, A04406). Subsequently, a dark incubation of 2 h with a goat anti-rabbit IgG secondary antibody (Abways Biotechnology, AB0101) took place. Finally, an infrared imaging scanning device (Odyssey LI-COR, USA) was utilized to visualize the results.

CCK8 assay

The T98G and U251 cell lines were inoculated into 96-well plates during the exponential growth phase, with a cell density of 5 × 103 cells per well (100 μl). Following transfection, cellular viability was evaluated over a four-day period at 0 d, 1 d, 2 d, 3 d, and 4 d post-transfection. To do so, 10 μl of CCK8 solution (Dojindo, Shanghai, China) was introduced into each well, and the optical density at 450 nm was measured two hours later using a SpectraMax Plus 384 microplate reader.

Colony assay

Cultured in six-well plates were glioma cells at a density of 1000 cells per well and incubated for a duration of two weeks. Following this, the cells underwent treatment with 4% paraformaldehyde and were stained with 0.1% crystal violet. Quantification of the colony numbers was conducted by analyzing the images with Image J software (version 1.52 2p).

5-Ethynyl-20-deoxyuridine (EdU) assay

Cells were plated at a density of 1 × 104 per well within a 96-well plate and left to incubate overnight. Following the initial incubation period, the cells underwent treatment with EdU (ShareBio, Shanghai, China) working solution for a duration of 2 h. Following this treatment, the cells were fixed utilizing 4% paraformaldehyde for a period of 15 min, followed by permeabilization with 0.1% TritonX-100 for a duration of 20 min. The Click-iT reaction mixture (ShareBio, Shanghai, China), which consisted of Azide, CuSO4, and reaction buffer supplement, was then introduced and left to incubate in darkness for approximately 30 min. Nuclei were subsequently stained with Hoeachst 33342 in total darkness for an additional 30-min duration. To conclude, the visualization of EdU-positive cells was accomplished utilizing an Axio Vert.A1 fluorescence microscope (Zeiss AG, Germany).

Transwell experiment

The research included two separate groups: Group 1 tested cell invasion ability with a matrix gel coating, while Group 2 tested cell migration ability without coating the matrix gel. In both groups, either U251 or T98G cells at a concentration of 1 × 105 were combined with 300 μl of medium lacking serum in the upper chamber, while the lower chamber contained 700 μl of medium with serum. Fixation and staining were performed following the identical method outlined previously after 24 h.

Wound healing experiment

T98G and U251 cells were initially plated into 6-well dishes and allowed to incubate for a duration of 24 h. A pipette tip was utilized to create slits by scratching the cell layer. Microscope images (Zeiss AG, Germany) were acquired at consistent positions both before and 24 h subsequent to scratching. To evaluate the cells' wound healing capacity, the scratch widths were measured at multiple time intervals using the Image J software.

Immunofluorescent staining

Glass coverslips were used to culture cells in 6-well plates. The cells were fixed with 4% paraformaldehyde, followed by treatment with 0.5% Triton X-100 (Solarbio, Beijing, CN) and blocking with 1% goat serum albumin. Subsequently, primary antibodies, including E-Cadherin (Abways Biotechnology, Shanghai, CN, #P12830), N-Cadherin (Abways Biotechnology, Shanghai, CN, #P19022), and Vimentin (Abways Biotechnology, Shanghai, CN, #P08670), were incubated overnight at 4°C. The cells were then exposed to the secondary antibody Goat Anti-Rabbit IgG H&L (Alexa Fluor-594, abcam #ab150080). DAPI (Servicebio, Wuhan, CN) was used for counterstaining for 10 min. Images were acquired with an Axio Vert.A1 fluorescence microscope (Zeiss AG, DE). Positive cells were quantified with Image J software, and fluorescence intensity was analyzed using Image-Pro Plus 6.0 software by an unbiased observer.

Statistical analysis

Analysis was performed with R version 4.1.3, 64-bit. Relevant R packages such as 'ggplot2', 'ggpubr', 'survival', and 'survminer', alongside other related R packages sourced from Bioconductor or R package repositories, were utilized. To evaluate prognostic values and compare patient survival among various subgroups within each dataset, Kaplan–Meier survival analysis and the log-rank test were implemented. The Wilcoxon test and Kruskal–Wallis test were employed for comparing two independent samples and multiple samples of non-parametric data, respectively. Prognostic variables were identified by screening clinical characteristics of high-risk and low-risk groups via univariable and multivariable Cox regression. Additionally, Spearman correlation analysis was conducted to assess correlation coefficients. In all statistical analyses, a P value below < 0.05 was considered statistically significant (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001).

Results

Genetic variation and expression of anoikis-related genes in glioma

A total of 516 genes associated with anoikis were acquired from the Genecards and Harmonizome portals (Supplementary Table 2). Subsequent differential gene expression analysis identified 219 differentially expressed genes (DEGs) from the TCGA-GTEx cohort and 303 DEGs from the GSE4290 cohort (Fig. 1A, B). A Venn diagram indicated 168 anoikis-related genes (ANRGs) (Fig. 2A). To construct a new cohort, the TCGA and GSE43378 datasets were merged to account for batch effects, resulting in the TCGA-GSE43378 cohort encompassing 17,662 genes. Univariate Cox regression analysis indicated that 153 of the 168 ANRGs had significant associations with survival (p < 0.05, km < 0.05). A forest plot highlighted that 28 of these ANRGs were highly correlated with glioma (p < 0.001) (Fig. 2B, Supplementary Table 3). Except for PDGFRA, CD24, MYC, and ANGPTL2, the majority of these 28 genes (24 in total) were linked to poor prognosis. A network diagram was used to elucidate the relationship between the expression levels of these genes (Fig. 2C). Given the frequent chromosomal abnormalities found in glioma patients, we also conducted an analysis of CNV data from the TCGA database to examine the chromosomal locations and alterations of these genes related to anoikis (Fig. 2D, E). Specifically, significant gains on chromosome 11 were observed for CD44 and KIF18A, while BRCA2 displayed a notable loss on chromosome 13.

Fig. 1.

displays the differential expression of genes related to anoikis (ANRG) in glioma. A The heat map depicting ANRG in the TCGA-GTEx dataset. B The heat map illustrating ANRG in the GSE4290 dataset. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Fig. 2.

displays the differential expression of genes related to anoikis (ANRG) in glioma. A A Venn diagram demonstrating the overlap of ANRGs in the TCGA-GTEx and GSE4290 datasets. B A forest diagram presenting the identification of the top 28 ANRGs (P < 0.001) through univariate COX regression analysis. C A network diagram showing the correlation among the top 28 ANRGs, with red and blue lines indicating positive and negative correlations, respectively. D Localization of the 28 ANRGs within chromosomal regions. E The frequency of CNV mutations of the 28 ANRGs in TCGA-GSE43378. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Consensus clustering of glioma molecular subpopulations using 28 anoikis-related genes

To explore the function of genes associated with cellular detachment-induced apoptosis (anoikis) in glioma, we utilized the Consensus Cluster Plus R software tool. Employing this software, we conducted consensus clustering based on the outcomes of 28 ANRGs linked to patient prognosis and univariate COX analysis (p-value less than 0.001). Through this analysis, the cohort was effectively segmented into two distinct subtypes, labeled as A and B, when the clustering parameter k was set at 2 (Fig. 3A). Examination of overall survival data revealed a notable prognostic distinction between the aforementioned subtypes (p-value below 0.001) (Fig. 3B). Further verification of the clustering accuracy was achieved through principal component analysis (PCA) (Fig. 3C). The findings indicated that subtype B, characterized by heightened ANRG expression, could potentially indicate a poorer prognosis for individuals with glioma. This observation was reinforced by the expression patterns of ANRGs and corresponding clinicopathological features of these subtypes (Fig. 3D). Additionally, we utilized the GSVA tool to carry out enrichment analysis of KEGG and GO pathways associated with differentially expressed genes, exploring the distribution of the 28 ANRGs in the distinct clusters (Fig. 3E; Supplementary Fig. 1). Noteworthy findings indicated that subtype B, linked with a dismal prognosis, was primarily involved in ECM receptor interactions and various oncogenic pathways. The significance of the extracellular matrix (ECM) in other cancer types has been well-documented. For example, ECM elevation has been observed in prostate cancer tissues [25, 26], contributing to tumor invasion and metastasis in gastric cancer [27]. In colorectal cancer, ECM facilitates the progression of epithelial-mesenchymal transition (EMT) in tumor cells [28], a critical mechanism for cancer cells to thrive in new environments.

Fig. 3.

Differentiating glioma subtypes linked to 28 ANRGs. A Consensus clustering results in a K = 2 matrix. B Comparison of survival times for A and B subgroups. C ANRG expression-based classification of subtypes using PCA. D Assessment of ANRG expression and clinicopathological traits for both subtypes using heat map. E Examination of KEGG pathway enrichment disparities between A and B subgroups through GSVA analysis. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Differential gene expression and immune infiltration between two subtypes

The use of box plots was employed to illustrate the patterns of expression of genes related to anoikis in the two subtypes. It is clear that PDGFRA, CD24, MYC, and ANGPTL2 show significantly lower levels of expression in subtype B compared to subtype A, while other genes related to anoikis exhibit elevated expression. These genes, which are differentially expressed and have an impact on overall survival, could be vital for predicting the prognosis of glioma patients and may be potential targets for precise therapy (Fig. 4A). Regarding the distinctions in the tumor microenvironment (TME) between the two subtypes, aside from monocytes, we noted increased levels of various immune cell infiltrates in subtype B relative to subtype A (Fig. 4B).

Fig. 4.

Patterns of immune infiltration and gene expression in the two subtypes. A ANRG expression in the two subtypes. B Immune infiltration patterns detected using ssGSEA in the two subtypes. (*P < 0.05, **P < 0.01, ***P < 0.001, ****i < 0.0001)

Construction and validation of loss-of-nest-related prognostic signatures with good performance

In order to assess the clinical significance of ANRGs, a study was conducted utilizing 28 ANRGs associated with survival (p < 0.001). These genes underwent Lasso-penalized Cox analysis (Fig. 5A, B). Following this, a multivariate Cox analysis was carried out to forecast the prognosis and survival rate of glioma patients using four ANRGs. A risk score was computed based on the expression levels of these four ANRGs and their corresponding coefficients (Supplementary Table 4). The formula for the risk score was determined as follows: Risk score = (0.255 * PDZ-binding kinase (PBK)expression level) + (0.182 * PLAU expression level) + (0.183 * Serine protease inhibitor clade E member 1 (SERPINE1) expression level)—(0.358 * adipokine angiopoietin-like protein 2 (ANGPTL2) expression level). Patients from the TCGA-GSE43378 cohort were divided randomly into training (n = 360) and test (n = 360) groups. Analysis of the KM curve indicated that higher risk scores were linked to poorer survival rates in both the training and test sets (Fig. 5C, D). The precision of the risk score model was assessed using the AUC of the time-dependent ROC curve. AUC values for the 1-year, 3-year, and 5-year training groups were 0.880, 0.890, and 0.870, respectively. Similarly, AUC values for the 1-year, 3-year, and 5-year test groups were 0.880, 0.890, and 0.870, respectively. These findings suggest that the risk score model can reliably anticipate survival outcomes (Fig. 5E, F). KM and ROC curves for all patients in the entire cohort generated similar results (Supplementary Fig. 2 and 3). Moreover, significant distinctions were noted in the risk scores between subtypes A and B (Fig. 6A). The alluvial plot illustrated the relationships among subtypes, risk scores, and life status (Fig. 6B).

Fig. 5.

Identification of prognostic features associated with Anoikis. A Using LASSO analysis with a tenfold cross-validation, 4 prognostic genes associated with Anoikis were identified. B Coefficient spectra for the 4 prognostic genes related to Anoikis. C Kaplan–Meier (KM) survival curves depicting prognoses for high-risk and low-risk groups in the training cohort. D Kaplan–Meier (KM) survival curves showing prognoses for high-risk and low-risk groups in the test cohort. E Time-dependent Receiver Operating Characteristic (ROC) curves of Overall Survival (OS) in the training cohort at 1, 3, and 5 years. F Time-dependent ROC curves of OS in the test cohort at 1, 3, and 5 years. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Fig. 6.

Immune microenvironment characteristics of glioma tissues across varying risk levels. A Risk score distributions for subtypes A and B. B Impact plot illustrating subtype, risk score, and survival status. C Relative proportions of immune cell infiltration among different risk categories. D Analysis of the correlation between risk score and M0 macrophage proportion in glioma tissues. E Analysis of the correlation between risk score and CD8 + T cell proportion in tumor samples. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Gene set enrichment analysis and immune activity with different risk scores

The immune microenvironment has a critical role in the development of tumors and the response to immunotherapy. In this investigation, we aimed to analyze the pattern of the TME in high-risk and low-risk groups of glioma patients. To quantify the proportions of infiltrating immune cells, we utilized the CIBERSORT R script. Initially, we arranged the glioma samples based on their risk scores, from low to high, to present the distribution of various immune cells (Fig. 6C). The levels of M0 macrophages (R = 0.53) and CD8 + T cells (R = 0.32) increased gradually as the risk scores increased (Fig. 6D, E). These results suggest that the presence of M0 macrophages and CD8 + T cells may play a significant role in the unfavorable prognosis in glioma patients. Furthermore, investigating the relationships between immune cells in glioma patients could provide valuable insights into the makeup of the immune microenvironment in specific tumor types (Fig. 7A, B). The four gene markers utilized to devise the ANRGscore model exhibited different expression patterns between the high-risk and low-risk groups and were closely linked to the infiltration of various immune cells (Fig. 7C, D). Moreover, we used ESTIMATE to evaluate the characteristics of infiltration in the tumor microenvironment of glioma patients. Interestingly, we observed that the StromalScore, ImmuneScore, and ESTIMATEScore were notably higher in the high-risk group compared to the low-risk group (p < 0.001) (Fig. 7E).

Fig. 7.

Immune microenvironment characteristics of glioma tissues across varying risk levels. A Comparison of immune cell compositions between high-risk and low-risk groups. B Correlation analysis between risk scores and the infiltration of tumor microenvironment (TME) cells. C Heatmap depicting expression patterns of model genes in both high- and low-risk categories. D Correlation analysis between immune cells and the four model-specific genes. E Calculated scores of expression patterns in high-risk versus low-risk groups. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Establishing a prognostic nomogram for glioma patients

Incorporating information on gender, age, tumor grade, and risk score into the nomogram (Fig. 8A) was crucial to considering the impact of clinicopathological characteristics on the prediction model. A calibration plot (Fig. 8B) was created to evaluate the agreement between the prognostic model's predicted overall survival (OS) and actual OS, which demonstrated the accuracy of the nomogram predictions. Analysis of the cumulative risk curve (Fig. 8C) revealed that patients with high glioma patient risk scores had an increasingly higher risk of overall survival. DCA, a method used to assess clinical prediction models, showed that the risk score was a reliable predictor of the survival rate in glioma patients at 1, 3, and 5-year intervals, along with clinical characteristics (Fig. 8D, E, F). A forest plot (Fig. 8G) indicated that the risk score and grade were the primary influencing factors. These findings highlighted that the ANRG-based risk score nomogram effectively predicted patient prognosis in clinical practice. By utilizing the ‘oncopredict’ R package, we found that chemotherapy drugs like gefitinib, erlotinib, lapatinib, and afatinib could be employed in glioma treatment for patients with high-risk scores, suggesting these patients may be more responsive to treatment (Fig. 8H) and impacting the selection of chemotherapy drugs for glioma patients.

Fig. 8.

Prognostic value of risk scores in individuals with glioma. A A nomogram integrating risk scores and clinicopathological factors. B Calibration plot utilized to verify the accuracy of the nomogram. C A cumulative hazard curve indicating the survival probability over a given period. D DCA curves depicting risk scores and clinical characteristics of glioma patients at the 1-year mark. E DCA curves showing risk scores and clinical characteristics of glioma patients at the 3-year mark. F DCA curves illustrating risk scores and clinical characteristics of glioma patients at the 5-year mark. G A forest plot presenting multivariable COX regression analysis of clinical characteristics along with risk scores in glioma patients. H Identification of four highly sensitive drugs (gefitinib, erlotinib, lapatinib, and afatinib) in patients categorized in the low-risk group. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Analysis of the correlation between ANRG and tumor immune microenvironment

The GSE131928 single-cell dataset from the TISCH database was used to analyze the expression of four ANRGs in the TME. This dataset includes 26 cell populations and 8 intermediate cell types, as illustrated in Fig. 9A. Malignant cells and macrophages mainly express PBK and PLAU, while ANGPTL2 is predominantly found in malignant cells and oligodendrocytes. SERPINE1, on the other hand, exhibits predominant expression in malignant cells (Fig. 9B, C).

Fig. 9.

ANRG expression in glioma TME-associated cells. A GSE131928 annotation for various cell types and their respective percentages. B, C PBK, PLAU, SERPINE1, and ANGPTL2 expression in individual cell types. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

In vitro studies experimentally verify the function of the key gene PLAU

The risk score model was constructed using four ANRGs, where the genes with the highest correlation coefficients were chosen for experimental validation. PBK, SERPINE1, and ANGPTL2 have been established in previous studies as promoters of glioma, whereas the role of PLAU in glioma remains uncertain. Despite the existing literature, we opted to investigate the impact of PLAU on glioma using cell lines. Analysis from the GEPIA database initially indicated a negative correlation between increased PLAU expression in glioma patients and prognosis (Fig. 10A). Subsequent in vitro experiments confirmed the functionality of PLAU. qRT-PCR results revealed upregulation of PLAU in all four glioma cell lines compared to the normal HA cell line, with T98G and U251 cell lines showing the highest expression levels (Fig. 10B). Following this, gene knockdown was carried out in T98G and U251 cell lines using si-PLAU3, which exhibited effective knockdown efficiency and was selected for further experiments (Fig. 10C). The knockdown of PLAU in both cell lines resulted in a significant decrease in PLAU expression, confirmed by Western blot analysis (Fig. 10D). Subsequent CCK-8 assays demonstrated a marked reduction in proliferation activity upon PLAU knockdown in T98G and U251 cell lines (Fig. 10E). Additionally, the colony formation ability of both cell lines was significantly impaired post-PLAU knockdown (Fig. 10F). Moreover, the EdU assay indicated suppressed proliferation of glioma cells following PLAU silencing (Fig. 10G).

Fig. 10.

Results from the in vitro experiment following PLAU knockdown are presented. A Survival analysis using the GEPIA database indicates that high PLAU expression correlates with poor glioma prognosis. Low PLAU cutoff is 180 months, High PLAU cutoff is 220 months. B PCR detection showed that PLAU expression is elevated in U87, T98G, A172, and U251 glioma cell lines, with the highest levels observed in T98G and U251 (n = 3). C siRNA transfection into T98G and U251 cell lines led to a significant decrease in the mRNA expression of PLAU (n = 3). D Similarly, siRNA transfection into T98G and U251 cell lines also substantially down-regulated the protein expression of PLAU (n = 3). E CCK-8 assays indicated a significant reduction in cell viability in T98G and U251 cell lines following PLAU knockdown (n = 3). F Colony formation assays showed a marked reduction in the cloning ability of T98G and U251 cells post-PLAU knockdown (n = 3). G Edu assays demonstrated a significant decline in the proliferation capacity of T98G and U251 cells after PLAU knockdown (n = 3). Scale bars: G 200 μm; (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

In order to examine the impact of PLAU on cell migration and invasion, we carried out transwell assays for both invasion and migration, along with a scratch wound healing assay. Results indicated a notable decrease in migration and invasion capabilities of T98G and U251 cell lines following PLAU knockdown in both transwell experiments (Fig. 11A). Moreover, T98G and U251 cell lines exhibited compromised healing abilities post PLAU knockdown (Fig. 11B). Given the importance of EMT in the invasive and metastatic processes of tumors, we explored the association between PLAU and the expression levels of EMT-related proteins (E-cadherin, N-cadherin, and Vimentin) in gliomas. Analysis via Western blotting indicated a significant rise in E-cadherin expression, coupled with reduced N-cadherin and Vimentin expression upon PLAU gene suppression in T98G and U251 cell lines (Fig. 11C). These findings were further supported by immunofluorescence staining, which illustrated an increase in E-cadherin expression and a decrease in N-cadherin and Vimentin expression in T98G and U251 cell lines post PLAU knockdown (Fig. 11D).

Fig. 11.

Results from the in vitro experiment following PLAU knockdown are presented. A Transwell experiment: Following PLAU suppression, there is a notable decrease in the migratory and invasive capabilities of T98G and U251 cell lines (n = 3). B Wound healing experiment: Subsequent to PLAU knockdown, the migratory abilities of T98G and U251 cell lines diminish significantly (n = 3). C Western blot Experiment: Alterations in the levels of EMT-associated proteins post PLAU knockdown (n = 3). D IF experiments indicate that suppression of PLAU impedes alterations in the expression levels of EMT-related proteins in T98G and U251 cell lines (n = 3). Scale bars: A and B: 200 μm; D:100um; (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

Discussion

Glioma, known as the most prevalent invasive tumor within the central nervous system [29], consistently presents challenges even with progress in surgical methods, radiotherapy, chemotherapy, and various other treatments [30]. The malignancy of gliomas results in a heightened invasion into normal tissues and spaces [31]. Therefore, developing a prediction model that utilizes anoikis-related genes could present a valuable strategy for early intervention. Nonetheless, the existing number of these markers is inadequate, emphasizing the urgent requirement for discovering additional biomarkers with superior predictive capabilities.

Anoikis is a particular type of programmed cell death that occurs when cells lose or inappropriately initiate cell adhesion. It is closely associated with tumor invasion and metastasis [32]. Anoikis takes place when cells detach from the ECM, leading to the apoptosis of misplaced or unattached cells [33]. This process is essential for maintaining tissue balance as it prevents shed cells from reattaching to new surfaces in incorrect positions [34]. However, unlike normal epithelial cells, cancer cells exhibit resistance to Anoikis and can survive and multiply without attaching to the ECM. This resistance plays a key role in the metastatic process, allowing cancer cells to thrive and move without proper ECM contact, enabling them to spread beyond the primary tumor [35]. The abnormal regulation of Anoikis has garnered significant attention in the scientific community owing to its correlation with anchorage-dependent growth and epithelial-to-mesenchymal transition, both crucial phases in tumor advancement and the metastatic dissemination of cancer cells [36]. For example, glioma displays Anoikis resistance, which facilitates its invasion of neighboring brain tissue. Numerous pathways are implicated in the development of anoikis resistance in glioma [37]. Research has revealed that the activation of heme oxygenase 1 by ATF4 impedes anoikis and fosters metastasis [38]. Furthermore, MNX1 has been demonstrated to decrease sensitivity to anoikis by activating TrkB in human glioma cells [39]. The examination of multiple genes demonstrates the complex interplay of various factors that impact anoikis resistance in tumor pathology. Thus, this encompassing method of scrutinizing multiple genes may yield valuable insights into tumor biology, bolstering clinical decision-making in the realm of cancer precision medicine.

In this research, we have identified a risk score signature comprising four genes: PBK, PLAU, SERPINE1, and ANGPTL2. Past studies have validated the strong correlation of these genes with tumor growth, invasion, and spread. For example, Hanlin Ma et al. showed that PBK boosts the aggressive characteristics of cervical cancer through the ERK/c-Myc signaling pathway [40]. The overexpression of PBK triggers the metastasis of hepatocellular carcinoma by activating the ETV4-uPAR signaling pathway [41]. Guangjin Chen et al. illustrated that PLAU boosts cell proliferation and epithelial-mesenchymal transition in head and neck squamous cell carcinoma [42]. Ting Yu discovered that METTL3 fosters colorectal cancer metastasis by enhancing the stability of PLAU mRNA in an m6A-dependent fashion [43]. Yanjun Gao et al. unveiled that PLAU is linked with cell migration and invasion, and its regulation is governed by the cervical cancer transcription factor YY1 [44]. Moreover, MircoRNA-1275 enhances the proliferation, invasion, and migration of glioma cells through SERPINE1 [45]. Treatment with antiplatelets elevates the secretion of SERPINE1, triggers MMP1 expression, and intensifies the metastasis of colon cancer [46]. Furthermore, ANGPTL2 heightens breast cancer by enhancing CXCR4 signaling, which results in bone metastasis of cells [47]. ANGPTL2 also advances the metastasis of hepatocellular carcinoma tumors [48].

In our research, we discovered differential gene expression between glioma and normal tissues. Utilizing unsupervised consensus clustering analysis, we categorized glioma patients into two distinct subgroups based on 28 ANRGs related to prognosis. Subsequent analysis and evaluation of these subgroups revealed disparities in survival duration. Moreover, GSVA indicated that the two clusters exhibited differences in immune cell infiltration and pathways related to metastasis. A risk score model based on four ANRGs was created and its prognostic accuracy was validated in both the training and test cohorts of glioma patients. This risk score model was strongly correlated with tumor development and immune cell infiltration. When examining clinical features such as patient gender, age, and tumor grade, the risk score was identified as a significant independent prognostic factor. This implies that the risk score is a reliable prognostic predictor, with higher scores associated with worse outcomes.

The TME is pivotal in both tumor metastasis and the efficacy of targeted therapies. We conducted an analysis to ascertain the proportions of 22 distinct immune cell types across various subtypes. Our findings highlighted that high-risk groups, which exhibited poor survival rates, had a significant increase in M0 macrophages and CD8 + T cells. This indicates a substantial role for these immune cells in glioma progression. Moreover, we found higher correlation coefficients between PLAU and SERPINE1, two genes included in the risk model, and M0 macrophages and CD8 + T cells.

PLAU, a vital gene present in the risk score signature we developed, is linked to unfavorable prognosis in glioma. Known as urokinase-type plasminogen activator (uPA), PLAU belongs to the PA family of serine proteases. Its primary role involves converting plasminogen to plasmin, breaking down proteins related to extracellular matrix remodeling, and activating growth factors [49]. Our experimental data reveals elevated levels of PLAU in glioma cell lines. Moreover, silencing PLAU leads to a significant decrease in the proliferation, invasion, and migration capacities of glioma cells, providing additional proof of its involvement in glioma progression. Research has indicated that PLAU is upregulated in various cancers, including head and neck squamous cell carcinoma [50], gastric adenocarcinoma [51], and breast cancer [52]. This indicates that PLAU plays a role in tumor advancement by facilitating cell proliferation, migration, invasion, and epithelial-mesenchymal transition. The above experimental results further bolster the conclusions drawn from our bioinformatics analysis.

Conclusion

The focus of this research was to distinguish between two subsets associated with anoikis in glioma and construct a predictive risk model. Our model exhibited robust and efficient capabilities in forecasting the survival outcome of glioma individuals. Moreover, the nomogram derived from this model could aid healthcare professionals in tailoring glioma therapy during medical procedures. These results bear considerable importance in investigating the molecular pathways and mechanisms of glioma, as well as enhancing therapeutic modalities and prognosis. Additionally, they could provide significant implications for precision healthcare.

Abbreviations

EMT

Epithelial-to-mesenchymal transition

PLAU

Plasminogen activator urokinase

CNS

Central nervous system

GBM

Glioblastoma

ECM

Extracellular matrix

TME

Tumor immune microenvironment

LGG

Low-grade glioma

GEO

Gene Expression Omnibus

GTEx

Genotype-Tissue Expression

UMAP

Uniform manifold approximation and projection

KEGG

Kyoto Encyclopedia of Genes and Genomes

GO

Gene Ontology

LASSO

Least absolute shrinkage and selection operator

ROC

Receiver operating characteristic

KM

Kaplan-Meier

TCGA

The Cancer Genome Atlas

GEO

Gene Expression Omnibus

DCA

Decision curve analysis

EdU

5-Ethynyl-20-deoxyuridine

Authors’ contributions

HT and ABZ contribute equally. ZMZ and LQL oversaw the overall design of this research. HT and ABZ performed the experiments, analyzed the data, and wrote the manuscript. WT, ABZ, RT contributed to the R software analysis and provided R language modification. YPS verified the data analysis. SYZ, DQ revised the figures and tables. ZFZ revised the discussion of the article. All authors contributed to the article and approved the submitted version.

Funding

Central Guiding Local Science and Technology Development Fund Projects (236Z7752G); the Medical Research Project of Hebei Provincial Health Commission (20230031); Special Project for the Construction of Hebei Province International Science and Technology Cooperation Base (193977143D).

Data availability

The data used to support the findings of this study are included in the article.

Declarations

Ethics approval and consent to participate

This study was approved by the Medical Ethics Committee of the Second Hospital of Hebei Medical University (No. 2024-R190) with the informed consent of all study participants.

The studies involving human participants were reviewed and approved by the Ethics Committee of the second Hospital of Hebei Medical University.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.