Background
Glioma is the most prevalent primary tumor of the central nervous system [
1]. It is classified into grades I to IV by the World Health Organization based on its malignant features, wherein grades I, II, and III are low-grade glioma and grade IV is high-grade glioma, also known as glioblastoma multiforme (GBM) [
1,
2]. GBM is the most malignant form of glioma with an extremely poor prognosis. Despite considerable advances in the development of treatments, including surgical resection, radiotherapy, and chemotherapy, little progress toward prolonged survival and better prognosis has been achieved over the last few decades [
3]. The modest median overall survival (OS) time in GBM is approximately 14 months, and only 5% to 6.8% of patients with GBM survive 5 years after diagnosis [
1‐
4]. Multiple clinical trials, including those on immunotherapy, have been conducted for patients with GBM; however, the results did not conclude the expected results [
1‐
4]. In our previous study, we generated a ferroptosis-related prognostic risk score model to predict the clinical significance and immunogenic characteristics of GBM [
5]. However, the biomarkers and predictors for patient outcome and the immunotherapy response of GBMs have not been fully elucidated, and existing predictive models are far from satisfactory.
Beyond classical apoptosis, several forms of regulated cell death (RCD) have been identified, such as ferroptosis, necroptosis and pyroptosis [
6]. These RCD subroutines differ in the initiating stimuli, intermediate activation events, and end effectors. Various heavy metals are essential micronutrients; however, the insufficiency or excessive abundance of metals can trigger cell death, which can induce RCD through different subroutines. For example, ferroptosis has been defined as an iron-dependent form of oxidative cell death caused by unrestricted lipid peroxidation [
7]. A novel RCD form of copper-induced cell death called “cuproptosis” was proposed by Tsvetkov et al. [
8], which is gaining attention in the field. Cuproptosis differs from oxidative stress-related cell death (e.g., apoptosis, ferroptosis, and necroptosis). In contrast, mitochondrial stress, especially the aggregation of lipoylated proteins and destabilization of Fe-S cluster proteins, results in proteotoxic stress and ultimately cell death. Hence, it may provide a new prospective for cancer treatment by regulating cuproptosis.
Methods
Data preparation
The omics data of the GBM samples from the TCGA database, including mRNA expression, Single Nucleotide Variant (SNV), copy number variation (CNA), and clinical information, were downloaded from UCSC Xena (
https://xenabrowser.net/). The mutation, CNA, and intracomatous heterogeneity of samples were derived from previous studies [
9]. The sample set from the CGGA (
http://www.cgga.org.cn/) database was used as an independent validation set, involving 139 (CGGA1-mRNAseq325) and 124 (CGGA2-mRNA-Array301) GBM samples. Data for scRNA-seq were downloaded from the GEO (
https://www.ncbi.nlm.nih.gov/geo/) database (GSE173278). Clinical characteristics of the three data sets were summarized (see Additional file
1: Table S1).
Cuproptosis activation scoring model
Ten cuproptosis characteristic genes (FDX1, LIAS, LIPTI, DLD, DLAT, PDHA1, PDHB, MTF1, GLS, and CDKN2A) were obtained from previous studies [
8]. Based on these characteristic genes, consistent clustering analysis was used to identify two sample clusters in the TCGA-GBM sample. The gene expression count data of two sample clusters were calculated based on R packet DESeq2 to identify differentially expressed genes (DEGs) [
10]. Candidate prognostic genes were identified from differentially expressed genes based on a univariate Cox regression analysis, and redundant factors were further filtered using LASSO Cox analysis. Based on the contribution of the prognostic genes to principal components 1 and 2, CuAS was defined as:
$$cuproptosis activation score (CuAS) = {Gene}_{HR > 1}* (PC1 + PC2)- {Gene}_{HR < 1} * (PC1 + PC2)$$
where HR (hazard ratio) is derived from the Cox analysis. The CuAS model in the validation cohort was reproduced with a similar formula. The construction of the cluster and CuAS models is illustrated in a schematic diagram (Fig.
14).
Biofunction prediction
The GO/KEGG enrichment analysis based on GSVA calculated the GO/KEGG signature activity scores in each sample, and the significant differences in the activity scores between the sample groups were compared. In addition, gene set enrichment analysis (GSEA) was used to calculate the differential expression of genes between the high and low CuAS groups, and the enrichment significance was calculated.
Overall survival outcome prediction
Samples were grouped into the high or low CuAS groups according to the median value of the CuAS. The OS difference between the high and low CuAS groups was predicted with the Kaplan–Meier algorithm. The receiver operating characteristic (ROC) curve and area under the curve (AUC) were generated to compare the prognostic ability within the different models.
GBM immune landscape
The immune and stroma scores and the tumor purity of the tumor samples were calculated based on the ESTIMATE algorithm [
11]. The cell composition of the tumor microenvironment (cellular infiltration) was calculated based on the CIBERSORT [
12] and xCell algorithms [
13] and GSVA score of the 28 immune cell signature genes, respectively [
14].
Tissue sampling from glioma patients
Fresh GBM tissues from histologically confirmed cases were obtained from the Union Hospital, Tongji Medical College, Huazhong University of Science and Technology. The study was approved by the ethics committee of the Union Hospital, Tongji Medical College, Huazhong University of Science and Technology.
Cell culture, real-time polymerase chain reaction, and immunohistochemistry
The normal human astrocyte cell line HA1800 and human glioma tumor cell lines U87, U251, LN229 and A172 were purchased from the Cell Bank of the Chinese Academy of Sciences. The STR identification reports of the cancer cell lines are presented in Additional materials (see Additional materials-cell lines STR identification), and we also used CCLA, an excellent cell line identification database, for secondary identification to ensure no cross-contamination of cell lines [
15]. The cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) (Gibco) containing 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin. qRT-PCR was conducted to compare the gene expression in 20 tumor samples in adjacent normal tissue. qRT-PCR was performed in triplicate using samples derived from three independent experiments. Formalin-fixed, paraffin-embedded GBM tissues were used for IHC staining. The primers’ sequence are provided (see Additional file
1: Table S3).
Lentivirus infection assay
The assay complies with the protocol described in a previous article [
16]. Short hairpin RNA (shRNA) against EREG (shEREG) and a negative control shRNA (sh NC) were designed and synthesized by GeneCreate (Wuhan, China). The lentivirus, pLent-shEREG-Flag-Puro or its negative control (NC) pLent-Flag-Puro (GeneCreate) was used to infect GBM cells with enhanced infection solution (GeneCreate) according to the manufacturer’s protocol. Seventy-two hours after the cells were infected with lentivirus, 2 μg/mL puromycin was added to kill the cells that had not been transfected. shRNA sequences are provided (see Additional file
1: Table S2).
Cell counting kit-8 assay
U87 and U251 cells were assessed with the CCK-8 (Biosharp, China) reagent according to the manufacturer’s instructions. Cells were inoculated on 96-well plates at a density of 2000 cells per well with 100 μL of medium. CCK8 solution (10 μL) was added to each well every 24 h for a total of 96 h, and the cells were further incubated at 37 °C for 1 h. The absorbance of each well was measured at 450 nm with a spectrophotometer.
U87 and U251 cells were prepared into a single cell suspension and seeded into a six-well plate (200 cells/well) for a two-week incubation to form colonies. After staining with 0.01% crystal violet (Sigma), the colonies were subjected to microscopic examination. The rate of colony formation was calculated.
Cell invasion and migration assays
After starving the cells for 6–8 h in serum-free DMEM, a total of 1 × 104 cells were seeded in the upper chamber with 200 μl of serum-free medium for the migration assay. In addition, 2 × 104 cells were added into Matrigel‑coated upper transwell chambers for the invasion assay. The lower chambers were filled with DMEM containing 10% FBS. After incubation at 37 °C for 24 h, cells on the lower surface of the membrane were fixed in 100% methanol and stained with 0.1% crystal violet dye for 20 min at room temperature. Finally, after washing with phosphate-buffered saline, the cells were imaged in five randomly selected fields under a light microscope (Olympus Corporation) at × 100 magnification.
5-ethynyl-2’-deoxyuridine (EdU) incorporation assay
According to the manufacturer’s instructions, the EdU Kit (Roche, Mannheim, Germany) was utilized to monitor the proliferation of transfected cells. A Zeiss Axiophot Photomicroscope (Carl Zeiss, Oberkochen, Germany) was used to capture representative images.
Compound
Elesclomol (STA-4783) was obtained from MedChemExpress (MCE). CuCl2 (Copper (II) chloride, 97%, 222011), and FeCl3 (reagent grade, 97%, 157740) were obtained from Sigma-Aldrich.
Western blotting
Proteins from tissues and cells were extracted using radioimmunoprecipitation assay (RIPA) (strong) buffer (Beyotime, Shanghai, China) containing protease inhibitors. Subsequently, protein concentrations were determined using a Bicinchoninic Acid (BCA) Protein Assay Kit (Beyotime). A total of 20 or 30 μg of protein was subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred to a 0.22 or 0.45 µm polyvinylidene difluoride (PVDF) membrane (EMD Millipore, Bedford, USA). PVDF membranes were then blocked in 5% skim milk for 2 h. Subsequently, samples were incubated with specific primary antibodies at 4 °C overnight. Following this, membranes were incubated with the appropriate secondary antibodies for 2 h at room temperature. Finally, the protein bands were visualized with enhanced chemiluminescence (ECL) Western blotting substrate (New Cell & Molecular Biotech). Information on the antibodies are provided (see Additional file
1: Table S4).
Flow cytometry cell cycle assay
After transient transfection, U87 and U251 cells were fixed in 75% ethanol for 12 h. Subsequently, cells were stained with propidium iodide (Beyotime) for cell cycle analysis. Finally, the percentage of cells in each cell cycle phase (G0/G1, S, and G2/M) was assessed, and the results were analyzed using the ModFit LT software.
RNA velocity and cells communication
The RNA velocity of the tumor cells was calculated using the package ‘velocity’ and ‘scVelo’ in Python. The various states of the GBM cells was mapped to show their internal transformation. The cross-talk between immunocytes and GBM cells was analyzed using the R package ‘celltalker,’ and differential ligand-receptor pairs were identified.
Transcription factor (TF) regulatory network construction
The RcisTarget human database was downloaded from
https://resources.aertslab.org/cistarget/ for transcription factor regulatory network construction. The corresponding gene ranking motif database (Hg38_refseq-r80_10kb_up_and_down_tss.mc9nr.feather, annotations_fname motifs-v9-nr.hgnc-m0.001–00.0.tbl) were downloaded from the human transcription factors list (https:/github.com/aertslab/pySCENIC/tree/master/resources), which is based on psSCENIC transcription factor regulation network. The AUCell algorithm was used to calculate the activity of each transcription factor, and the regulation module was identified according to the Connection Specificity Index (CSI). The calculation method of CSI was based on a previous article [
17]. Similarly, we used the hTFtarget database to predict between TF and targets, which contains the most comprehensive data on human TF-target to date [
18]. The overall activity score of the regulatory module was defined as the mean of all TF activities in the module.
Prediction of potential drug sensitivity
The drug sensitivity information and corresponding expression level were obtained from Genomics of Drug Sensitivity in Cancer (GDSC), Cancer Cell Line Encyclopedia (CCLE), and the Cancer Therapeutics Response Portal (CTRP) (
https://portals.broadinstitute.org/ctrp). The CuAS score of each cell line was calculated and grouped based on the median. The correlation between the AUC and IC50 data of multiple drugs in the cell lines was calculated by using Spearman’s correlation. The difference of the AUC value between the two groups were compared by the Wilcoxon test.
Statistical analysis
The significance of the difference between the two groups of continuous variables was evaluated using the Wilcox rank-sum test. Spearman’s rank correlation was used to evaluate the correlation between the variables. Univariate and multivariate Cox regression and LASSO Cox regression were used to identify molecules with prognostic efficacy, and the K-M curves and log-rank tests were used to assess the survival differences between the sample groups. All computational analyses were performed by R (version 4.1.2) or Python.
Discussion
Copper is an essential cofactor in all organisms; however, it is toxic for cells when concentrations of copper exceed thresholds maintained by an evolutionarily conserved homeostasis mechanism [
19,
20]. In fact, it is not known how excessive copper can induce cell death. However, the Broad Institute has currently identified a new mechanism that is different from known cell death: cuproptosis [
8]. Cuproptosis is a kind of cell death that is dependent on mitochondrial respiration. Copper directly binds to lipoylated components of the tricarboxylic acid cycle. Afterwards, aggregation of these copper-bound, lipoylated mitochondrial proteins and subsequent Fe-S cluster protein loss trigger proteotoxic stress and a distinct form of cell death [
19‐
22]. Cuproptosis is involved in cell death, and the Broad Institute paper suggests that drugs that inhibit mitochondrial respiration may be a strategy against disease [
19‐
22]. In addition, many mitochondrial proteins have a high degree of respiration function in various cancer cells [
23]. Thus, copper ion metal carriers may be a new method for cancer treatment.
To the best of our knowledge, this study was the first paper to comprehensively analyze the association between copper-induced cell death and GBM by combining scRNA-seq and bulk RNA-seq data. First, we identified two sample subgroups based on the characteristic genes of cuproptosis. We found that immune checkpoint genes (PD-1, IDO1 and LAG3) and cancer hallmarks (fatty acid metabolism, KRAS, P53, NOTCH, and PI3K/AKT/MTOR signaling pathway) showed significant differences between the two subgroups. Immune checkpoint is a kind of immunosuppressive molecule, which can regulate the intensity and breadth of the immune response, to avoid the damage and destruction of normal tissues. In the process of tumor occurrence and development, immune checkpoint has become one of the main reasons for immune tolerance. Subsequently, we constructed CuAS based on the differential genes of subgroups, which contained 11 genes, including 8 coding genes and 3 non-coding genes. EREG was the gene with the largest contribution coefficient to the principal component, so we focused on EREG. EREG is a 19-kDa peptide hormone that belongs to the Epidermal Growth Factor (EGF) family of peptide hormones [
24]. Epiregulin binds to the EGF receptor (EGFR/ErbB1) and ErbB4 (HER4) and stimulates signaling of ErbB2 (HER2/Neu) and ErbB3 (HER3) through ligand-induced heterodimerization with a cognate receptor [
24]. EREG possesses a range of functions in both normal physiologic states as well as in pathologic conditions. EREG contributes to inflammation, wound healing, tissue repair, and oocyte maturation by regulating angiogenesis and vascular remodeling and by stimulating cell proliferation [
24]. Deregulated EREG activity appears to contribute to the progression of a number of different malignancies, including cancers of the bladder, stomach, colon, breast, lung, head and neck, and liver [
2,
7,
24]. EREG is also associated with imaging omics as an important prognostic gene and MRI parameters revealed that hemodynamic abnormalities were associated with the expression level of the mTOR‐EGFR pathway in patients with GBM [
25]. Rab27b promotes the proliferation of adjacent cells and radio-resistance of highly malignant GBM cells through EREG-mediated paracrine signaling after irradiation [
26]. Furthermore, EREG activates the extracellular signaling-related kinase/MAPK pathway in GBM, suggesting that the inhibition of the EREG-EGFR interaction may be a strategy for EREG-overexpressing patients with GBM [
2]. In our study, we detected EREG mRNA expression and protein levels in tissues and multiple glioma cell lines. IHC staining revealed that the EREG protein expression in tumors was higher than that in normal tissues; the result of WB also showed similar results. Knockdown of EREG can inhibit the proliferation, invasion, and migration of tumor cells. EGFR and PDL1 expression of protein were down-regulated after knockdowning of EREG. Moreover, we explored if EREG could influence the process of cuproptosis. Cell vitality assay demonstrated that only the coexistence of Cucl
2 and ES can influence the cell vitality and that other metals had no effect. The effect of ES-Cu required a specific concentration range (5 nM-50 nM). shEREG can revert the cell vitality that is influenced by cuproptosis. Therefore, we detected the protein expression of FDX1 in the shEREG and shNC groups. The results showed that FDX1, the core regulatory protein in cuproptosis, was down-regulated in the shEREG group.
Combined with the single cell transcriptome, the model of cuproptosis was analyzed, and the GBM sample cells were divided into seven types, including three types of malignant cells (OLIG1 + malignant, VEGFA + malignant, and CENPF + malignant). OLIG1 and other oligodendrocyte markers were highly expressed in OLIG1 + malignant cells, which may be oligodendrocyte progenitor glioma mother cells. VEGFA, CHI3L1, and other angiogenesis related markers were highly expressed in VEGFA + malignant cells, which may have a strong ability to induce local angiogenesis and may be associated with invasion/metastasis. CENPF + , TOP2A, UBE2C, and other markers are associated with the cell cycle and may be mesenchymal glioma blasts, which may be associated with tumor proliferation/invasion [
27,
28]. Others types observed were microglia, fibroblasts, endothelial cells, and oligodendrocytes. High CuAS was found in VEGFA + malignant cells. Based on CNV [
29], OLIG1 + malignant cells were the ancestor clones. The function of VEGFA + malignant cells demonstrated that the pathways were mainly enriched in those related to hypoxia and stress, which is also consistent with the fact that cuproptosis is mitochondrion-dependent programmed cell death. Activated cells with high CuAS scores based on differences between high and low CuAS transcription factors were concentrated in the VEGFA + malignant cell subpopulation, reflecting the potential association between CuAS scores and Module1. The VEGF and CD99 signaling pathways were significantly enriched in high CuAS cells. VEGF specifically binds to Fltl and KDR/Flkl on the surface of endothelial cells, resulting in a variety of biological effects [
30]. VEGF is closely associated with angiogenesis and development [
31]. VEGF plays an important role in all stages of tumor formation, inducing the production of a large number of proteolytic enzymes, reducing the basement membrane of the host blood vessels, weakening the barrier effect, increasing the permeability of blood vessels, promoting a large amount of fibrinogen exudation, and forming a new matrix necessary for tumor adhesion and migration [
30,
31]. Angiogenesis is determined by the growth and metastasis of solid tumors. VEGF degrades extracellular matrix by inducing endothelial cells to express protease, resulting in metastasis, proliferation, and angiogenesis [
32]. CD99 is abnormally expressed in many different types of tumors, and plays an important role in the diagnosis, development, metastasis, and prognosis, mainly affecting the invasion and metastasis of tumor cells [
33]. Immunofluorescent detection of tissue samples with high and low CuAS showed that VEGFA and CD99 were also highly expressed in tissues with high CuAS, and the results were opposite in tissues with low CuAS, which provided a new idea for us to intervene in cuproptosis-related tumor cells.
Immunotherapy is essential in tumor treatment. Despite the lack of specific immune cohort verification for glioma, several other tumor immune cohorts have shown the possibility of treatment for patients with high CuAS. Considering that EREG may affect the expression of PDL1 and the immune process, we believe that immunotherapy may have therapeutic opportunities for patients with high CuAS. Chemotherapy is also the first line of treatment for glioma. We predicted the potential targeted drugs for high CuAS GBM cells. Methotrexate can be used to treat GBM owing to several factors such as the upregulation of CD73 [
34]. Pharmacological inhibition of DNA-PKcs with the DNA-PKcs inhibitor NU7441 reduced GSC tumorsphere formation [
35] mTORC1/2 inhibitors of KU - 0063794 can inhibit PI3K-Akt-mTOR signaling in glioblastoma and reduce cell proliferation [
36]. The PI3K inhibitor GDC-0941 enhances radio-sensitization and reduces chemo-resistance to temozolomide in GBM cell lines [
37]. Cabozantinib is a potent, multitarget inhibitor of MET and VEGF receptor 2 [
38]. NVP-BEZ235 (PI3K and mTOR a dual inhibitor) can inhibit the PI3K pathway to hinder glycolytic metabolism in GBM cells [
39].
However, there were some limitations of our study. First, cuproptosis is a new concept, and there are few characteristic genes of cuproptosis, so it may affect the stability and applicability of the model on single-cell data. Second, the VEGF and CD99 signaling pathways were only detected by immunofluorescence, and further experiments are needed to prove their correlation with cuproptosis. Third, we found that EREG is closely related to PDL1 and FDX1, but further direct mechanisms are needed to reveal the relationship between them.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.