Clinical features and molecular landscape of cuproptosis signature‐related molecular subtype in gastric cancer

Abstract Recent studies have highlighted the biological significance of cuproptosis in disease occurrence and development. However, it remains unclear whether cuproptosis signaling also has potential impacts on tumor initiation and prognosis of gastric cancer (GC). In this study, 16 cuproptosis‐related genes (CRGs) transcriptional profiles were harnessed to perform the regularized latent variable model‐based clustering in GC. A cuproptosis signature risk scoring (CSRS) scheme, based on a weighted sum of principle components of the CRGs, was used to evaluate the prognosis and risk of individual tumors of GC. Four distinct cuproptosis signature‐based clusters, characterized by differential expression patterns of CRGs, were identified among 1136 GC samples across three independent databases. The four clusters were also associated with different clinical outcomes and tumor immune contexture. Based on the CSRS, GC patients can be divided into CSRS‐High and CSRS‐Low subtypes. We found that DBT, MTF1, and ATP7A were significantly elevated in the CSRS‐High subtype, while SLC31A1, GCSH, LIAS, DLAT, FDX1, DLD, and PDHA1 were increased in the CSRS‐Low subtype. Patients with CSRS‐Low score were characterized by prolonged survival time. Further analysis indicated that CSRS‐Low score also correlated with greater tumor mutation burden (TMB) and higher mutation rates of significantly mutated genes (SMG) in GC. In addition, the CSRS‐High subtype harbored more significantly amplified focal regions related to tumorigenesis (3q27.1, 12p12.1, 11q13.3, etc.) than the CSRS‐Low tumors. Drug sensitivity analyses revealed the potential compounds for the treatment of gastric cancer with CSRS‐High score, which were experimentally validated using GC cells. This study highlights that cuproptosis signature‐based subtyping is significantly associated with different clinical features and molecular landscape of GC. Quantitative evaluation of the CSRS of individual tumors will strengthen our understanding of the occurrence and development of cuproptosis and the treatment progress of GC.


INTRODUCTION
Gastric cancer (GC) is one of the most important cancers in the world, with the fifth incidence rate and the fourth mortality rate of the global world.The incidence rate is highest in east Asia and Eastern Europe [1].The risk factors for gastric cancer prognosis include helicobacter pylori infection, age, tumor stage, subtype, and so forth [2].With the increase of the aging population, the number of new cases of gastric cancer will continue to increase in the foreseeable future.In recent years, attempts have been made to combine the molecular types of gastric cancer with histological phenotypes and clinical features to understand the mechanism of occurrence and development of gastric cancer and explore important indicators to guide clinical decision making [3][4][5].The exploitation and identification of new biomarkers have promoted the development of systemic therapy and targeted therapy [6][7][8].At present, the primary treatment regimen of gastric cancer is surgical resection combined with perioperative adjuvant therapy [9].There is growing appreciation that the GC patients with programmed death ligand 1 (PD-L1) positive would benefit from the immune checkpoint inhibitor therapy [10].Moreover, recent studies reported that vascular endothelial growth factor (VEGF) receptor inhibitor as a single agent or in combination with chemotherapy have improved the patients' survival outcomes [11].At present, the early predictive screening and postoperative monitoring markers of gastric cancer are not clear.Prognosis biomarkers should be developed and utilized for the stratification of high-risk population of gastric cancer to guide the therapy and intervention of gastric cancer.The personalized precision medicine services for patients using big data technology and standardized statistical models [12,13].
Copper, an indispensable trace element for the human body, can exert cytotoxic effects and trigger programmed cell death when present in excessive amounts [14].Previous studies have shown that both copper ionophores (DSF, etc.) and copper chelators (TTM, etc.) are considered to have anticancer effects [15][16][17].Recent research indicates that copper ionophores trigger a particular type of controlled cellular demise, one that operates through mechanisms distinct from those of other forms of programmed cell death, such as apoptosis, pyroptosis, necrosis, and ferroptosis.[18].Copper-dependent cell death arises from the direct interaction of copper with lipoylated components of the tricarboxylic acid (TCA) cycle [18].This interaction results in the aggregation of lipoylated proteins and the subsequent depletion of iron-sulfur cluster proteins, ultimately leading to proteotoxic stress and the induction of cell death [18].Two groups, positive hits (FDX1, LIAS, LIPT1, DLD, DLAT, PDHA1, PDHB) and negative hits (MTF1, GLS, CDKN2A), were divided according to the molecular function of cuproptosis-related genes [18].Copper enrichment was found in both serum and tumor tissue from patients with a variety of tumors [19][20][21].At present, the research of cuproptosis in liver cancer, lung cancer, triple negative breast cancer and other tumors has been gradually carried out [22][23][24][25].Prognostic models based on cuproptosis characteristics in various tumors are mostly constructed from one aspect, with few integrating clinical features, molecular landscapes, and genetic mutation features to construct models, and there is a lack of validation through molecular biology experiments [26][27][28][29][30]. Research in the field of liver cancer has made rapid progress, with studies constructing prognostic models and verifying that copper deathrelated gene LIPT1 may promote the proliferation, invasion, and migration ability of liver cancer cells [31].Other studies have demonstrated that copper can mediate copper-dependent cell death to suppress the proliferation of breast cancer cells and decrease tumor volume before breast cancer surgery through the development of a hydrogel system [32].In the field of breast cancer, the latest research found that copper sensitization system therapy can target and induce copper poisoning programs in internal and external tumor cells, and significantly inhibit lung metastasis of breast cancer [33].In gastric cancer, the construction of prognosis models related to copper mortality features is not complete, and there is a lack of prognosis models based on the clinical characteristics and detailed molecular landscape features of gastric cancer patients.However, it remains to be explored in the development of molecular-targeted therapy for gastric cancer.
In this study, we delved into the interactions among cuproptosis-related genes, considering them as a cuproptosis-related signature to investigate their association with prognosis and molecular pathways in GC.A novel index termed as "cuproptosis signature risk score (CSRS)," which based on the principle components of the cuproptosis-related transcriptomic, was utilized for evaluation of cuproptosis risk and survival conditions.We conducted an analysis of the genetic mutation landscape associated with cuproptosis in distinct risk subtypes of GC patients, revealing marked differences between the two subtypes.Our predictive model holds promise for improved prognosis prediction in patients, carrying significant clinical implications.

The landscape of the mutation of cuproptosis-related genes in gastric cancer
In this study, we investigated the roles of 16 cuproptosisrelated genes in GC.GO enrichment and Metascape analyses were performed on 16 cuproptosis-related genes, revealing significantly enriched biological processes that are summarized in Figure 1A.These genes were mainly enriched in glyoxylate metabolism, glycine degradation, copper metabolism, and so forth.Genetic alterations, predominantly missense mutations, insertional mutations, and frameshift mutations, were observed in 81 of 408 (19.85%) samples of cuproptosis-related genes.The most common base mutations are C to T, T to C, and C to A (Figure 1B).Among the cuproptosis-related genes, CDKN2A exhibited the highest mutation frequency, closely followed by ATP7B and ATP7A.We further investigated the co-occurrence of mutations among all cuproptosis-related genes and identified significant mutation co-occurrence relationships between CDKN2A and ATP7A, ATP7B and LIPT1, ATP7A and ATP7B, as well as ATP7A and LIPT1 (Figure S1A and Table S1).Upon further analysis of the 16 cuproptosis-related genes, it was evident that CNV mutations were prevalent.Figure 1C illustrates the chromosomal locations of CNV alterations for these genes.Additionally, our analysis indicated that the frequencies of CNVs across the 16 cuproptosis-related genes were inconsistent, suggesting variable rates of mutation across the gene set.CDKN2A and LIPT1 had prevalent CNV deletions, whereas the ATP7B, DLD, MTF1, GLS, and ATP7A showed widespread CNV amplification (Figure 1D).
Furthermore, to assess the mutual regulation among the cuproptosis-related genes, Spearman correlation analysis was conducted.The results revealed a significant negative correlation between DBT and ATP7A with other cuproptosis-related genes.Conversely, DLAT, FDX1, and PDHA1 demonstrated a significant positive correlation with other genes associated with cuproptosis (Figure S1B).Cox regression analysis was employed to determine the association between cuproptosis-related genes and the prognosis of gastric cancer (GC) patients.The forest plot generated from this analysis indicated that DBT, ATP7A, PDHB, and CDKN2A could be regarded as risk factors significantly linked to overall survival in GC patients (Figure S1C).The interaction relationships between 16 cuproptosis-related genes were also visualized in the form of a relationship network, whose lines representing the correlation between the two genes and circle size representing the prognosis value (Figure 1E).The intricate wiring highlighted 16 cuproptosis-related genes acting together to influence the development and progression of gastric cancer and also provided a rationale for subtyping based on cuproptosis signature.Average gene essentiality scores, specifically CRISPR-Cas9 gene knockout scores (CERES), were computed across 26 cell lines originating from stomach adenocarcinoma (STAD).These scores serve as a metric to assess the degree of gene dependence in these cell lines.FDX1, LIAS, DLST, GCSH, and DLD in the majority of cell lines were below −0.25 score and had a potential impact on cell proliferation (Figure 1F).The expression patterns of cuproptosis-related genes differed notably between tumor specimens and adjacent normal specimens.Specifically, genes such as LIPT1, ATP7A, LIAS, DBT, and PDHB were found to be significantly downregulated in the tumor tissue.Conversely, genes like CDKN2A, GLS, MTF1, and SLC31A1 exhibited marked upregulation in the tumor, suggesting distinct roles in the pathophysiology of the disease (Figure 1G).

Cuproptosis signature patterns characterized by specific clinical features and molecular subtypes
The relationships between the expression patterns of cuproptosis-related genes and clinical characteristics, as well as molecular subtypes, were further investigated in GC tumors.Based on the 16 cuproptosis-related signature genes, we obtained four stable transcriptomic clusters by unsupervised clustering analysis (Figure S2A,B).We observed notable differences in the expression patterns of cuproptosis-related genes across distinct clusters, indicating potential variations in the underlying biological processes and mechanisms associated with these genes in different subpopulations of gastric cancer.LIAS, DLAT, PDHA1, and SLC31A1 were significantly elevated in the CSC1 subtype; CDKN2A and GLS were markedly increased in the CSC2 subtype; MTF1, DBT, and SLC31A1 were evidently increased in the CSC3 subtype; ATP7A and LIPT1 were obviously increased in the CSC4 subtype (Figure 2A, ACRG cohort).A similar expression distribution of the cuproptosis-related genes was also observed in The Cancer Genome Atlas (TCGA) and Yonsei cohort (Figure S2C,D).Interestingly, a preponderance of microsatellite instability (MSI)-positive tumors converged within the CSC1 subtype, culminating in the amalgamation of a hypermutated phenotype, MSI, and an immune-enriched subtype.Conversely, the CSC4 subtype exhibited a strong correlation with advanced tumor staging, the mesenchymal phenotype, the diffuse histological subtype, and a fibrotic tumor microenvironment (TME) subtype.These observations underscore the intricate relationship between genetic instability, tumor microenvironment, and gastric cancer progression (Figure 2B,C), which indicated stromal invasion and worse prognosis.Further survival analysis uncovered marked prognostic disparities amongst the four clusters defined by cuproptosis signatures in gastric cancer (GC) samples.Notably, the CSC1 signature demonstrated a favorable prognostic correlation, whereas the CSC4 signature was associated with adverse survival outcomes (Figure 2D-F and Table S2).Multivariate Cox proportional hazards regression analysis further confirmed that this stratification model exhibited an independent association with patient survival outcomes (ACRG cohort: CSC1 vs. CSC4, hazards ratio [HR], 2.24 [95% CI, 1.43-3.50],p < 0.001, Figure S2E; TCGA cohort: CSC1 vs. CSC4, HR, 2.31 [95% CI, 1.32-4.02],p = 0.003, Figure S2F; Yonsei cohort: CSC1 vs. CSC4, HR, 1.85 [95% CI, 1.08-3.19],p = 0.026, Figure S2G;).Additionally, we examined the distribution in cuproptosis scores obtained through single sample gene set enrichment analysis (ssGSEA) across different clusters.Our analysis revealed significant differences in the distribution of cuproptosis scores among clusters in all three datasets.Specifically, the scores for CSC3 and CSC4 were found to be significantly lower compared to CSC1 and CSC2 (Figure S2H-J).

Cuproptosis signature clusters characterized by distinct molecular processes and genomic alterations
To investigate the underlying biological molecular changes of the four clusters based on cuproptosis signatures, we conducted gene set variation analysis (GSVA) enrichment analysis using the Kyoto Encyclopedia of Genes and Genomes (KEGG) gene set.The results of GSVA revealed that CSC1 exhibited significant enrichment in processes related to substance synthesis, metabolism, and energy metabolism, including aminoacyl-tRNA biosynthesis, one-carbon pool by folate, glyoxylate and dicarboxylate metabolism, arginine and proline metabolism, citrate cycle (TCA cycle), and oxidative phosphorylation.Conversely, CSC4 predominantly showed enrichment in pathways associated with cell adhesion and stromal pathways, such as glycosaminoglycan biosynthesis (chondroitin sulfate), regulation of actin cytoskeleton, focal adhesion, and ECM-receptor interaction.These findings suggest distinct biological characteristics and potential functional roles of different clusters in gastric cancer.Interestingly, all of these relevant biological processes were median enriched in CSC2 subtype (Figures 3A and S3A).Moreover, we utilized the immuno-oncology signatures, which curated from Zeng et al. study [34], to evaluate the biological process and pathway across four CSC subtypes.CSC1 and CSC2 had a higher citric acid cycle score in the 3 databases (Figures 3B and S3B,F); CSC1 had a higher lipoic acid metabolism in the 3 databases (Figures 3B and S3C,G); CSC4 had a lower ferroptosis score in ACRG cohort (Figure 3B), and CSC1 had a higher ferroptosis score in TCGA cohort (Figure S3E); CSC1 and CSC2 had a higher pyruvate metabolism score (Figures 3B and S3D,H); Glyoxylate and dicarboxylate metabolism score was decreased progressively from CSC1 to CSC4 in Yonsei cohort (Figure S3I).To gain deeper insights into the mutational landscape across distinct CSC subtypes, we conducted SMG analysis within the TCGA cohort, comparing gene mutation frequencies among CSC1, CSC2, CSC3, and CSC4 subsets.Our analysis revealed that in the CSC1 subtype, ARID1A, PIK3CA, COL11A1, and APC exhibited higher mutation rates compared to the other subtypes within the TCGA cohort.These findings suggest potential subtype-specific mutational patterns that may contribute to the unique characteristics and behaviors of CSC1 in gastric cancer (adjusted chi-square test, p < 0.05; Figure 3C).We also explored the tumor microenvironment alterations in different CSC subtypes.Genomic analysis revealed that CSC1 exhibited the highest mutational load and CSC3 had a higher aneuploidy score (Figure 3D).Antitumor lymphocyte cell subpopulations, such as activated CD4 + /CD8 + T cells were mainly enriched in the CSC1 subtypes.However, effector memory CD4 + / CD8 + T cells, mast cells, eosinophils, Type 1 helper cells, and Type 2 T helper cells were markedly elevated in the CSC4 subtype (Figure 3E).

Construction of cuproptosis signature risk score and exploration of its clinical relevance
Despite our findings highlighting the prognostic significance of cuproptosis signature-based clusters, these analyses were limited to the patient population level and lacked precision in predicting cuproptosis signature patterns within individual tumors.To address this limitation, we devised a scoring system, termed the cuproptosis signature risk score (detailed in the Method section), aimed at quantifying the cuproptosis signature in individual GC patients.This scoring scheme enables a more personalized assessment of cuproptosis-related risk, facilitating tailored therapeutic approaches in gastric cancer management.In principal component analysis and the 16 cuproptosis-related genes contributing to the top 5 principal components, the top-ranked genes include DLAT, FDX1, DBT, DLD, and so forth (Figure S4A).We further divided the GC patients into CSRS-Low and CSRS-High subtypes.The cutoff point was identified by using standardized maximally selected log-rank statistics in ACRG cohort (Figure 4A).We examined the relationship between known biological signatures and the cuproptosis signature risk score through Spearman analysis.The resulting heatmap of the correlation matrix revealed notable correlations.Specifically, the cuproptosis signature risk score was markedly negatively correlated with lipoic acid, pyruvate metabolism, and the citric acid cycle.Conversely, it showed a positive correlation with signatures related to cancer fibroblasts, tumor-associated macrophages, glycogen and glycosaminoglycan biosynthesis, as well as cytokine receptors.These findings suggest complex interactions between the cuproptosisrelated risk and various biological processes in gastric cancer, pointing to potential targets for further investigation and therapeutic intervention (Figure 4B and Table S3).Our observations indicate a noteworthy difference in the expression levels of cuproptosis-related genes between the CSRS-Low and CSRS-High subtypes.DBT, MTF1, and ATP7A were significantly elevated in the CSRS-High subtype; ATP7B, SLC31A1, GCSH, LIAS, DLAT, FDX1, DLD, and PDHA1 were obviously increased in the CSRS-Low subtype in the 3 databases (Figures 4C  and S4B,C).We also explored the distribution between the cuproptosis signature risk scores and CSC clusters and found that the cuproptosis signature risk score was significantly increased in CSC4 subtypes than the other 3 subtypes among the three databases (Figure 4D-F).Considering the intricacies involved in quantifying the cuproptosis signature, we have employed an alluvial diagram to illustrate the workflow of constructing the cuproptosis signature risk score (Figure 4G).By presenting this workflow in a clear and concise manner, the alluvial diagram aids in understanding the methodology behind the cuproptosis signature risk score and facilitates the replication of our approach in future studies.To further validate the prognostic value of the cuproptosis signature risk score, we stratified patients into two subtypes: CSRS-Low and CSRS-High.As hypothesized, patients classified as CSRS-Low demonstrated a significantly better prognosis across all three cohorts studied.This finding underscores the potential of the cuproptosis signature risk score as a predictor of survival outcome in gastric cancer patients (Figure 4H-J and Table S2).Multivariate analysis of the three cohorts has indeed confirmed that the cuproptosis signature risk scores can serve as an independent prognostic biomarker in GC (HR, 1.61 [95% CI, 1.16-2.25],p = 0.005, ACRG cohort in Figure S4D; HR, 1.73 [95% CI, 1.19-2.50],p = 0.004, TCGA cohort in Figure S4E; HR, 1.49 [95% CI, 1.12-2.00],p = 0.007, Yonsei cohort in Figure S4F).

Tumor genomic landscapes in cuproptosis signature mutation gastric cancer
To gain deeper understanding of the genomic alterations associated with the cuproptosis signature risk score in gastric cancer patients, we analyzed somatic mutation data obtained from whole-exome sequencing (WES-seq) and single-nucleotide polymorphism (SNP) array analysis.Specifically, we conducted a significantly mutated gene (SMG) analysis in the TCGA cohort, comparing the frequency of gene mutations between the CSRS-Low and CSRS-High subtypes.This comprehensive approach allowed us to identify genetic aberrations that may underlie different risk scores and potentially contribute to the pathogenesis and progression of gastric cancer.The mutational profile in the TCGA cohorts showed that ARID1A, PIK3CA, APC, ERBB3, COL11A1, RNF43, BCOR, PTEN, and PTPN23 had higher mutation rates in the CSRS-Low subtype than in the CSRS-High subtype (adjusted chi-square test, p < 0.05, Figure 5A).A consistent result was also observed in the TCGA cohort, lower cuproptosis signature risk scores in patients were significantly associated with higher tumor mutational load (Figure 5B).After parsing the somatic mutation data from whole-exome sequencing (WES-seq) and single-nucleotide polymorphism (SNP) array analysis, we proceeded to analyze the single-nucleotide variants (SNVs) in gastric cancer (GC) tumors.Specifically, we compared the SNV profiles of CSRS-Low and CSRS-High subtypes across a matrix of 96 possible mutations.The pie chart shows that the CSRS-Low subtype had a slight decrease in C > T transitions and the CSRS-Low subtype had a slight increase in T > G transitions.The Lego plot analysis reveals distinct mutational patterns in gastric cancer (GC), with C > T transitions predominantly occurring at ApCpN trinucleotide sites.Notably, a specific mutation, T > G transition at GpTpG sites, is highlighted in the CSRS-Low subtype (Figure S5A).Subsequently, we extracted five mutational signatures from the genomic data (Figures S5B  and 5C), including defects in polymers POLE (COSMIC 10), spontaneous or enzymatic deamination of 5-methylcytosine (COSMIC 1), defects in the DNA-DSB repair by HR (COSMIC 3), defective DNA mismatch repair (COSMIC 26) and an unknown one (COSMIC 17) (Figure 5C).We found that CSRS-High subtype had higher mutational counts in signature 1 and signature 26; however, lower mutational counts in signature 3 and signature 10 (p < 0.05, Figure 5D).The arm-level somatic copy number alteration (SCNA) results suggest that specific chromosomal regions, including chr1p, chr3, chr7q, chr8q, chr11q, chr18q, and chr20p, contain the most frequently amplified or deleted regions in GC.These findings are significant because they implicate these chromosomal regions in the pathogenesis and progression of GC.Amplifications and deletions in these regions may lead to the dysregulation of oncogenes or tumor suppressor genes, respectively, driving tumor growth and metastasis (Figure 5E).The top-ranked CNV sites in two CSRS subtypes were 8q, 9p, 20q, and so forth (Figure S5D).The focal level SCNAs revealed that the cytobands in 1p36.23 (PARK7) in CSRS-Low subtype, and 3q27.1 (PIK3CA), 11q13.3(CCND1), and 12p12.1 (KRAS) in cuproptosis signature risk high score subtype contained the markedly amplified focal regions; and cytobands in 1p36.11(ARID1A) and 11q23.2(FDX1, DLAT, BACE1) in cuproptosis signature risk high score subtype contained the frequently deleted regions (FDR < 0.05, Figure 5F).Mutation counts attributed to the COSMIC 1 and COSMIC 26 signatures exhibited a notable elevation in the CSRS-Low subtype, while a conspicuous reduction in mutation counts linked to the COSMIC 3 signature was observed in the CSRS-High subtype (Figure S5C).

Analysis of correlation and effectiveness between CSRS score and antineoplastic drugs
We further jointly analyzed the Cancer Cell Line Encyclopedia (CCLE) and Genomics of Drug Sensitivity in Cancer (GDSC1) databases to determine the association between CSRS score and antineoplastic drug sensitivity of gastric cancer cell lines (Figure 6A and Table S4).

Verifying the differences between CSRS-Low (AGS) and CSRS-High (HGC-27) subtypes through Western blot experiments and drug sensitivity experiments
We conducted molecular experiments to validate our findings, and the results of Western blot analysis revealed notable disparities in the expression levels of FDX1, LIAS, DLAT, DLST, GCSH, PDHB and PDHA1 in gastric cancer cells between the CSRS-Low (AGS) and CSRS-High (HGC-27) subtypes.In the CSRS-High subtype of gastric cancer cells, the expression of proteins related to cuproptosis progression was significantly increased, while the CSRS-Low subtype was opposite (Figure 7A).In the CSRS-Low and CSRS-High subtypes, the expression of copper death-related proteins showed significant changes in some molecules after drug stimulation, and the specific mechanism needs further exploration.All three drugs can effectively inhibit the proliferation of two GC cell lines tested, and all three drugs have good inhibitory effects.Additionally, NSC319726, talazoparib, and thapsigargin have relatively high IC50s in the CSRS-Low subtype of gastric cancer cells (Figure 7B-D).The inhibitory effect of the drugs on the growth of gastric cancer cells was further confirmed by colony-forming assays (Figure 7E-H).

DISCUSSION
An increasing number of studies have shown that the cellular cuproptosis mechanism plays a great role in tumor development and progression [35].The utilization of multi-omics technology has enabled the identification of cancer biomarkers generated by tumor cells within the intricate tumor microenvironment, thereby significantly advancing the development of novel diagnostic and therapeutic approaches [36].There have been studies constructing prognosis models based on copper death features, but they have not comprehensively considered clinical features, and the explanation of the genetic molecular landscape is not in-depth enough [37].There is no cross-validation between drug screening and cell line sensitivity, and no sensitivity-related experiments have been conducted to verify the screened drugs [38].A column chart model combining risk scores and other clinical pathological features has been constructed based on copper death features and tumor microenvironment single cell profiles, but its application and screening of effective drugs based on clinical features and molecular landscapes of gastric cancer are not detailed enough [39].
In contrast to this study, which solely utilized the TCGA data set for analysis, our study takes a more comprehensive approach by incorporating three different datasets.Additionally, our study employs a different modeling approach (PC), which may provide unique insights and perspectives on the topic at hand [40].Part of these studies have explored the relationship between cuproptosis mechanisms and tumor genomics, but have not characterized the genomic landscape of specific tumors exhaustively.By considering multiple datasets and adopting a different methodology, our study aims to provide a more comprehensive and nuanced understanding of the subject matter.In this paper, we show that the cuproptosis signature is significantly associated with molecular landscapes and clinical characteristics of gastric cancer, which will strengthen our understanding of occurrence and development of copper-mediated cytotoxicity and treatment schedule of gastric cancer.In this study, utilizing an integrative clustering algorithm with cuproptosis-related genes, we delineated four distinct cuproptosis signature subtypes.Notably, among these genes, GLS encodes glutaminase, a pivotal enzyme catalyzing the conversion of glutamine to glutamate.Glutamine metabolism is a crucial process in cancer cell proliferation and viability, highlighting the potential significance of GLS in this context [41].FDX1 is a gene that encodes the protein ferredoxin 1, which is involved in electron transfer reactions in the mitochondria.Abnormal FDX1 expression has been associated with colorectal cancer and can affect cellular energy metabolism and redox balance [42].SLC31A1 is a gene that encodes the copper transporter protein CTR1.ATP7B, and ATP7A are genes that encode copper-transporting ATPases involved in copper homeostasis.Dysregulation of SLC31A1, ATP7B, and ATP7A has been implicated in colorectal cancer and can affect copper homeostasis and tumor growth [43,44].MTF1 is a gene that encodes the metal-responsive transcription factor 1, which plays a role in regulating the expression of genes involved in metal ion homeostasis and oxidative stress response [45].It can affect cellular response to oxidative stress and DNA damage.DBT, LIPT1, CDKN2A, PDHB, DLD, DLAT, LIAS, and DLST genes are involved in various cellular processes, including energy metabolism, DNA repair, and cell cycle regulation [35].They have been implicated in colorectal cancer and can contribute to tumor growth and progression.These 16 cup-associated genes differ in characteristics and mechanisms.The four subtypes had different molecular characteristics, immunological phenotypes, biological features, genomic features, and clinical prognosis in GC.CSC1 is mainly characterized by MSI molecular status, which corresponds to the immune enrichment phenotype.It is mainly enriched in the biological functions related to TCA cycle, pyruvate, and galactose metabolism et al.CSC4 is mainly characterized by epithelial-mesenchymal transition (EMT) molecular typing of gastric cancer, corresponding to the fibrous matrix phenotype and enriched in adhesion-related biological functions.CSC2 and CSC3 were featured by intermediate conditions from CSC1 to CSC4.Previous studies have shown that the classification of tumor microenvironments can help us identify patients who may benefit from combined target molecular therapy [46,47].Based on these results, we speculated that GC patients with the CSC4 pattern may benefit from combination treatment with the blockade targets on adhesion-related signaling pathways, and CSC1 patterns may benefit from blockade targets on tumor metabolism signaling.The results of immune infiltration showed a higher degree of T-cell infiltration in CSC1 and a higher degree of B-cell infiltration in CSC4.T cells are important for immune responses against pathogens and tumors.However, in colorectal cancer, their function can be suppressed, allowing the tumor to evade the immune system.Enhancing T cell function could improve colorectal cancer therapies.B cells, responsible for antibody responses, have a minor role in colorectal cancer.The role of B cells is relatively minor as this type of cancer typically does not trigger a strong antibody response.However, B cells may still contribute to the tumor microenvironment by modulating the immune response and promoting tumor growth [48].The mechanism was consistent with the prognosis of CSC1 and CSC4.By exploring the differences in the immune microenvironment of different subtypes of tumors, we can provide theoretical support for immune-targeted therapy for gastric cancer.
We further established a copper signature risk scoring system to evaluate patient prognosis and biological pathways and thus guide patient precise treatment.Further analysis revealed that patients with a higher CSRS score were mainly derived from CSC3 and CSC4 subtypes and had a worse prognosis.In the ACRG and Yonsei queues, the differences between CSC1 and CSC2, CSC3, and CSC4 are relatively significant.In the TCGA queue, there was no significant difference between the CSC1 and CSC2 groups, but a more significant difference compared to CSC3 and CSC4.The reason for this result may be that the TCGA cohort is mainly composed of Western populations, while the ACRG and Yonsei cohorts are mainly composed of Asian populations.Genomic analysis indicated that high CSRS subtype was characterized by decreased TMB and defective DNA mismatch repair signature.CNV with amplification of oncogenes, such as PIK3CA, KRAS, and CCND1, was found mainly focused in the CSRS-High subtype; whereas the cuproptosis inhibitors, such as FDX1 and DLAT, were markedly deleted in CSRS-High subtype.These evidence suggest that the CSRS was associated with genomic alteration, and higher score indicated the increasing genetic heterogeneity.CRC patients who have a high TMB may exhibit increased tumor immunogenicity, making it more likely for tumor cells to be recognized and attacked by the immune system.Consequently, CRC patients with high TMB may display heightened sensitivity to immunotherapy and are more likely to benefit from immune checkpoint inhibitor therapy [49].Mutations in the PIK3CA and KRAS genes are commonly found in colorectal cancer and are associated with increased tumor risk.PIK3CA mutations activate the PI3K signaling pathway, promoting tumor cell growth and survival.KRAS mutations lead to sustained activation of the KRAS protein, enhancing tumor cell growth and spread.Both mutations are associated with tumor aggressiveness, recurrence, and drug resistance in colorectal cancer [50].Additionally, several chemical regimes mentioned above were found to be sensitive to high CSRS scores, indicating their potential significance in therapy.
Notably, this study found that certain cuproptosisrelated genes play important regulatory roles in the biological processes of tumors.Recent studies have found that the higher expression of copper transporter ATP7A can promote tumorigenesis and metastasis, and is significantly related to the poor prognosis of breast cancer patients [51].Meanwhile, ATP7A can limit autophagymediated degradation of VEGFR2 and promote VEGFR signaling pathway and neoangiogenesis [52].This study showed similar results, in which higher ATP7A enrichment was observed in the CSRS-High GC samples.Previous studies have shown that DLAT was expressed in different gastric cancer types in different amounts, and found that knockdown of DLAT in DLAT high expression gastric cancer cell lines can affect the proliferative function of cells and may promote aerobic glycolysis pathway, oxidative phosphorylation and catabolic reactions [53].PM2.5 exposure augmented the expression of the glycolytic gene DLAT, thereby stimulating the glycolytic process and ultimately facilitating the development of NSCLC [54].Our results also showed that DLAT was mainly enriched in oxidative phosphorylation, citric acid cycle, pyruvate metabolism and other substances and energy metabolic processes in gastric cancer.DLAT-related mechanisms are relatively complex and need to continue to be explored.Some studies have found that FDXR with FDX2 deletion regulates P73 tumor suppressor via IRP2 to modulate aging and tumor suppression [55].FDXR exerts its tumor suppressive effect by interacting with P53 [56].FDX1 is closely related to glucose metabolism, amino acid metabolism and fatty acid oxidation and promotes ATP production in lung cancer, but it does not affect lung cancer cell proliferation and has no effect on apoptosis and cycle [57].FDX1 exhibits reduced expression in various cancer types and exhibits significant correlations with clinical parameters, TMB, MSI, and immune-associated pathways [58].Our study reveals that FDX1 is intimately linked to energy and substrate metabolism in gastric cancer, predominantly observed in patients with the CSRS-Low subtype and associated with a more favorable prognosis.Collectively, these findings implicate that a holistic evaluation of cuproptosis attributes could bolster our comprehension of the underlying physiological mechanisms involved in cuproptosis.
Interestingly, our correlation and sensitivity analyses for antineoplastic drugs revealed several drugs with both correlation and sensitivity the same trend, such as SB505124, thapsigargin, and NSC319726.SB505124 is a selective TGFβR inhibitors acting on ALK4 and ALK5, which can induce cell death and inhibit biological functions such as tumor cell invasion, proliferation, and survival through a variety of mechanisms [59,60].Thapsigargin induces apoptosis in almost all cells, but recent studies have shown that it can inhibit Notch1 signaling pathway, and its benefits need to be comprehensively evaluated [59,61].NSC319726 can bind copper to induce oxidative stress leading to cell cycle arrest [62].This can help us screen drug treatment targets for gastric cancer, and provide a theoretical basis for the selection of clinical precise treatment.And we preliminarily verified that the above three drugs have obvious effects on cell viability through molecular biology experiments, and then we need to verify the effects of the three drugs on cell proliferation, migration and invasion, cell cycle, apoptosis and other functions through experiments.Obviously, we need more evidence to support the clinical benefits of drugs.

CONCLUSION
In this study, we reviewed the literature and collated 16 cuproptosis-related gene signatures and identified that the cuproptosis stratification system was associated with different prognosis and molecular patterns.Although the results showed that model reliability and applicability were satisfactory, prospective clinical cohort studies are needed to validate the robustness of our model for clinical theragnostic, and subsequent studies focusing on biological experiments to validate its molecular mechanism in gastric cancer were needed to optimize the CSRS score system.Quantitative assessment of the CSRS in individual tumors will enhance our comprehension of cuproptosis occurrence, development, and treatment progression in GC.Encouragingly, cuproptosis-based translational medicine exhibits potential as a clinical candidate, pending rigorous safety and efficacy evaluations in human cancer trials.

Public data acquisition and data cleaning
We conducted a literature review focused on cuproptosis and compiled a list of 16 recognized cuproptosis-related genes.These genes were then carefully analyzed to identify distinct clusters associated with the cuproptosis signature.The 16 cuproptosis-related genes were divided into two groups, positive hits (FDX1, DLST, LIAS, LIPT1, DLD, DLAT, DBT, SLC31A1, PDHB, PDHA1, GCSH) and negative hits (MTF1, GLS, CDKN2A, ATP7A, ATP7B) (Table S5).Gastric cancer sample data with available transcriptomic and clinical data were obtained from public databases, including the ACRG data set (Asian Cancer Research Group, 2015, N = 298, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62254), TCGA-STAD data set (The Cancer Genome Atlas-Stomach Adenocarcinoma, 2018, N = 408, https://portal.gdc.cancer.gov/)and Yonsei data set (Yonsei University cohort, 2018, N = 433, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84437).A total of 1139 gastric cancer patients were included in the study for follow-up analysis after removing the samples with incomplete information.The ACRG data set was employed for model establishment, and TCGA and Yonsei were used for further validation.

Functional enrichment, interaction network analysis and pathway annotation
The Metascape database (https://metascape.org)was used for functional enrichment analysis of cuproptosisrelated genes.The protein-protein interaction network of 16 cuproptosis-related genes was using the STRING (https://string-db.org) to plot, which was based on gene expression levels.The enrichment pathway of the selected genes was based on GO enrichment analysis.

Copy number variation (CNV) analysis, high frequency mutant genes and tumor mutation characteristics
The R package "oncoplot" was used to perform the gene mutation landscape descriptions.Mapping of 16 cuproptosis-related genes to chromosomes was performed with the R package "RCircos."CNV analysis was performed with the R package "maftools" to annotate amplifications and deletions cytobands in different CSRS subtype.The somatic copy number alterations (SCNA) and aneuploidy scores for each gastric cancer sample were compiled from previous research studies [63].The Signature Enrichment function of Bayesian Inference for Nonnegative Matrix Factor Deconvolution Models was used to determine the optimal number of mutational signatures that could be extracted.Cosine similarity analysis was utilized in the Catalogue of Somatic Mutations in Cancer to compare and annotate the explored mutational signatures of gastric cancer.

Unsupervised cluster analysis of 16 cuproptosis-related genes
The iClusterPlus package implements a regularized latent variable model for clustering that incorporates joint inference across different data types to produce an integrated cluster assignment.This approach leverages information from multiple sources to improve the accuracy and robustness of clustering results.Here, we employed the iClusterPlus package to identify different clusters of cuproptosis signatures based on transcriptomic profile of 16 cuproptosis-related genes in GC.According to the iClusterPlus guidelines, the optimal number of clusters was determined to be four, as indicated by the leveling off of the curve representing the percentage of explained variation.This suggests that further increasing the number of clusters beyond four would not significantly improve the amount of variation explained by the model.We utilized the iCluster2 function with K = 4, lambda set to 0.1, and method set to "lasso" to perform the clustering analysis.

Establishment of the cuproptosis score via ssGSEA
We utilized a gene set comprised of cuproptosis genes and applied the ssGSEA method to assign a cuproptosis score to each sample.ssGSEA is a gene set-based approach that evaluates the enrichment of a gene set within an individual sample.It quantifies gene set enrichment by calculating the rank order of genes within each gene set, resulting in a gene set enrichment score [64].The key advantage of ssGSEA is its ability to assess gene set enrichment within a single sample, eliminating the need for inter-sample comparisons.This is particularly valuable when working with a limited number of samples or lacking a control group.ssGSEA can be applied to diverse types of gene expression data, including RNA-seq and microarray data.

Immune cell infiltration estimation with ssGSEA
The relative infiltration of 28 immune cell types in the GC tumor microenvironment were quantified by the ssGSEA.Special feature gene panels for each immune cell subset were curated from Charoentong et al. research [65].The relative abundance of each immune cell type was represented by an enrichment score in the ssGSEA analysis.The bio-similarity of the immune cell filtration was estimated by multidimensional scaling (MDS) and a Gaussian fitting model.

Cell lines and drug sensitivity analysis
We also evaluated the genetic vulnerabilities of the cuproptosis-related proteins in the GC using data from the Cancer Dependency Map Project (DepMap).Average gene essentiality scores (CRISPR-Cas9 gene knockout scores [CERES]) that reflect cuproptosis gene dependence were calculated in 26 cell lines with GC origin.Drug IC50 value, which was released by Genomics of Drug Sensitivity in Cancer (GDSC1) datasets, was utilized to explore the potential therapeutic association with CSRS scoring [66].

Construction of the CSRS scoring system
We constructed a CSRS scheme to quantify the relative cuproptosis risk level of individual patients by using principal component analysis (PCA).Specifically, the eigenvalues and eigenvectors of the correlation coefficient matrix on 16 cuproptosis genes were derived and subjected to construct matrix product (PC score).About 79.1% of the variation in PCA analysis is explained by this first five eigenvalues and considered as the number of principal components to retain.Then, the sum of the first five eigenvalues weighted by PC score were served as the cuproptosis signature score.We also adopted a formula to define the CSRS scheme of each GC sample: CSRS = ∑(E i *PC i )/E 1:5 , where E denotes the Eigen values, PC denotes the principal components, i denotes number from 1 to 5.Moreover, the surv-cutpoint function from R package "survival," which based on standardized maximally selected log-rank statistics, was used to determine the optimal point of stratification of high-risk and low-risk groups in the CSRS scoring system.The identified optimal cut-point of CSRS-score in ACRG was further utilized in TCGA and Yonsei cohort.

Western blot analysis
Observed that the cell density reaches about 90%, we prepared to lyse the cells to extract the protein.The protein concentration was measured by BCA method using a BCA protein assay kit (Solarbio; PC0020).After obtaining the OD value, the final concentration of each sample was calculated according to the concentration formula of standard curve calculation.Prepared a 12% concentration of the gel, we started protein electrophoresis after loading samples.First the protein was pressed in the homogenous bands in the concentrated gel with a constant pressure of 50 V, and then we run the separation gel with 100 V.After electrophoresis, we started to transfer the PVDF membrane, used a constant current of 220 mA, and calculated the transfer time according to 1 min/kDa.Skimmed milk was selected for blocking, and 50 rpm decolorization shaker was used for 1 h.After the PVDF membrane was washed with TBS-t solution, the primary antibodies targeting FDX1 (Proteintech; 12592-1-AP; 19 kDa), LIAS (Proteintech; 11577-1-AP, 34-42 kDa), DLD (Proteintech; 16431-1-AP; 54 kDa), DLAT (Proteintech; 13426-1-AP; 70 kDa), DLST (CST; 5556S; 50 kDa), PDHA1 (Proteintech; 18068-1-AP; 43 kDa), PDHB (Proteintech; 14744-1-AP; 39 kDa), GCSH (Proteintech; 16726-1-AP; 19 kDa), and β-actin (Proteintech; 20536-1-AP; 42 kDa) were added for incubation, and the 50 rpm decolorization shaker was shaken with ice box at 4°C for 14-16 h.After washied the PVDF membrane with TBS-t solution, we added the corresponding secondary antibody (Proteintech; SA00001-2) for antibody incubation, and decolorized the shaker at 50 rpm for 1 h.After washed the PVDF membrane with TBS-t solution, we prepared the developer, added an appropriate amount of developer to each strip, exposed and developed the protein strip, and analyzed the strip with ImageJ1.52vsoftware after storage.

Plate colony
We inoculated gastric cancer cells into a six-well plate, with 1000 cells per well.After 24 h of cell growth, we replaced the solution with a drug that measured the concentration of IC50.After 24 h of treatment, we switched to complete culture medium and continued to culture for 8-10 days until the cell colony size was suitable.After washing with the precooled PBS, we fixed it with 4% paraformaldehyde 1 mL for 30 min, dyed it with crystal violet 1 mL for 30 min, and washed it with PBS.Then we took photos with a camera, and counted the number of cell colonies using ImageJ1.52 v software.

Statistical analyses
The statistical analyses in this study were generated by R-4.0.2.For quantitative data, statistical significance for normally distributed variables was estimated by student's t tests, and non-normally distributed variables were analyzed by the Wilcoxon rank-sum test.For comparisons of more than two groups, Kruskal-Wallis tests and oneway analysis of variance were used as nonparametric and parametric methods, respectively.Chi-square test and Fisher's exact test were used to analyze contingency tables depending on specific grouping conditions.Kaplan-Meier survival analysis and the Cox proportional hazards model were used to analyze the association between the metabolic transcriptomic pattern and prognosis with the R package "Survminer" (0.4.6).The surv-cutpoint function from the "survival" package was applied to stratify samples into high and low CSRS-score subtypes in ACRG cohort.All comparisons were two-sided with an alpha level of 0.05, and the Benjamini-Hochberg method was applied to control the false discovery rate (FDR) for multiple hypothesis testing.

AUTHOR CONTRIBUTIONS
Wei Chong and Hao Chen developed the methodology, created models, programmed the computer code, and reviewed the published work.Huicheng Ren performed the experiments, wrote the initial draft, and revised the manuscript.Kang Xu did the experiments, collected the data, and revised the manuscript.Xingyu Zhu, Yuan Liu, Yaodong Sang, Han Li, Jin Liu, and Chunshui Ye collected the data, analyzed the study data, and annotated the study data.Liang Shang and Changqing Jing managed the responsibility for the research activity planning and execution.Leping Li formulated the overarching research goals and aims, gave the financial support, and provided the study materials.All authors have read the final manuscript and approved it for publication.

F I G U R E 1
Comprehensive characteristics of cuproptosis-related genes in gastric cancer.(A) Metascape enrichment network visualization showed that the intra-cluster and intercluster similarities of enriched terms of 16 copper-mediated cytotoxicity-related genes.Different clusters were coded in different colors.(B) 81 of the 408 gastric cancer (GC) patients underwent genetic alterations on 16 cuproptosis-related genes, with a frequency of 19.85%, mostly consisting of missense mutations, nonsense mutations, and insertion mutations.The number on the right panel indicated the mutation frequency of each cuproptosis-related gene.The distribution below the chart shows the mutation frequency of the single base group.Each column represented single patients.(C) The location of copy number variation (CNV) alteration of cuproptosis-related genes on chromosomes.(D) The CNV gain or loss of 16 cuproptosis-related genes in GC.The loss of CNV was labeled with pink dot; the gain of CNV was labeled with blue dot.(E) The correlation and prognosis of 16 cuproptosis-related genes in GC.The cuproptosis-related genes in different functions were depicted by circles in different colors.The lines linking cuproptosis-related genes represented their interaction with each other.The size of each circle represented the prognosis effect of each gene with scaled log-p value.Protective factors for patients' survival were indicated by a green dot in the circle center and risk factors indicated by the black dot in the circle center.(F) The influence of knockout of 16 cuproptosis-related genes on the proliferation in gastric cancer cell lines.The lower the CERES score, the greater the effect after the gene knockout.(G) The difference in mRNA expression levels of 16 cuproptosis-related genes between paired normal and GC samples.The asterisks represented the statistical p value (*p < 0.05; **p < 0.01; ***p < 0.001).

3
Cuproptosis signature patterns characterized by specific clinical features and molecular subtypes.(A) Heatmap shows the scaled expression of representative 16 cuproptosis-related genes curated from the Asian Cancer Research Group (ACRG) in distinct cuproptosis signature clusters.The clinical features and subtypes, including age, gender, stage, molecular subtype, Epstein-Barr virus (EBV), and microsatellite instability (MSI) were annotated with different colors.(B, C) Comparison of cuproptosis signature clusters with different molecular subtypes of gastric cancer in ACRG (B) and The Cancer Genome Atlas (TCGA) (C) cohort.(D-F) Kaplan-Meier curves of overall survival (OS) for 298 gastric cancer (GC) patients in ACRG cohort (D), 405 GC patients in TCGA cohort (E), and 433 GC patients in Yonsei cohort (F) with different cuproptosis signature clusters (log-rank test).The cuproptosis signature clusters characterized by distinct functional signal enrichment and immune landscapes.(A) Heatmap shows the enrichment signal pathways of Kyoto Encyclopedia of Genes and Genomes (KEGG) in four different cuproptosis signature clusters across three datasets.(B) Distribution of molecular signatures curated from Zeng et al. study, including citric acid cycle, ferroptosis, lipoic acid metabolism and pyruvate metabolism among CSC1, CSC2, CSC3, and CSC4 subtypes in Asian Cancer Research Group (ACRG) cohort.(C) The mutational landscape of significantly mutated genes (SMGs) of The Cancer Genome Atlas (TCGA) gastric cancer in four cuproptosis signature clusters.This waterfall plot depicts the mutation frequency of SMGs across four clusters, and genes with statistically significant differential distribution are highlighted in upper right asterisk.(D) Tumor mutation load and aneuploidy score among CSC1, CSC2, CSC3, and CSC4 subtypes were compared in ACRG cohort.(E) Comparison of the fraction of single sample gene set enrichment analysis (ssGSEA) algorithm annotated cell subsets in four cuproptosis signature clusters.Within each group, the thick line represented the median value.The bottom and top of the boxes were the 25th and 75th percentiles (interquartile range).The whiskers encompassed 1.5 times the interquartile range.The statistical difference of four clusters was compared through the Kruskal-Wallis H test. *p < 0.05; **p < 0.01; ***p < 0.001.
Construction of the cuproptosis signature risk score and exploration of its clinical relevance.(A) The cutoff point of the cuproptosis signature risk score identified by using the standardized maximally selected log-rank statistics.(B) Correlations between cuproptosis signature risk score (CSRS) score and the known biological gene signatures using Spearman analysis.The negative correlation was marked with blue and positive correlation with red.Number located in the circle indicated the correlation coefficients.(C) Heatmap shows the scaled expression level of cuproptosis-related genes and CSRC score among four CSC subtypes and two CSRS subtypes.Each column represented single patients.(D-F) Distribution of cuproptosis signature risk scores among four cuproptosis signature clusters in three independent datasets of Asian Cancer Research Group (ACRG) (D), The Cancer Genome Atlas (TCGA) (E), and Yonsei (F).(G) Alluvial diagram of cuproptosis clusters in groups with different molecular subtypes (MSS/TP53−, MSS/TP53+, microsatellite instability [MSI], epithelial-mesenchymal transition [EMT]), cuproptosis signature clusters (CSC1-CSC4), and cuproptosis signature risk scores subtype (Low and High).(H-J) Kaplan-Meier curves for high and low CSRS patient groups in the three independent datasets of ACRG (H), TCGA (I), and Yonsei (J).Log-rank test, p < 0.05.

F I G U R E 5
Tumor genomic landscapes in different cuproptosis signature risk score (CSRS) subtype of gastric cancer.(A) The mutational landscape of significantly mutated genes (SMGs) in The Cancer Genome Atlas (TCGA)-stomach adenocarcinoma (STAD) stratified by low (left panel) versus high CSRS (right panel) subtypes.Individual patients were represented in each column.Mutational frequencies of SMGs in different CSRS subtypes were depicted in two sides of the panel.Genes were highlighted in blue for those statistically significant by Fisher's exact test.*p < 0.05; **p < 0.01; ***p < 0.001.Age, gender, stage, molecular subtype, immune subtype, MFP and CSC were shown as patient annotations in the upper panel.(B) Comparison of tumor mutation load in CSRS-High versus low subtypes.(C) The mutational activities of the corresponding extracted mutational signatures (Catalog of Somatic Mutations in Cancer [COSMIC] 10, 17, 1, 3, and 26, named as COSMIC database).(D) Comparison of mutational activities of corresponding extracted mutational signatures in high versus low CSRS subtypes.(E) Arm-level somatic copy-number alteration (SCNA) events in high versus low CSRS subtypes.Red denotes amplification and blue denotes deletion.(F) Focal peaks with significant somatic copy-number amplification (red) and deletions (blue) (Q values < 0.1) are shown.The top 20 amplified and deleted cytobands are labeled.Representative genes encoded from these focal peaks are highlighted in approximate positions across the genome.