Somatic alteration landscape of cuproptosis regulators
To understand the genomic alterations in the 10 cuproptosis regulators in tumors, we analyzed the single nucleotide variation (SNV) and copy number variation (CNV) data from 10 680 pan-cancer samples to calculate the mutation frequencies and somatic copy number alterations (SCNA). Somatic alterations were defined as mutations or SCNA. Overall, the frequency of somatic alterations was very low for all cuproptosis regulators (1–3%), except CDKN2A that showed somatic alterations in up to 18% of the tumors (Fig. 1A). Most somatic alterations in CDKN2A were deep deletions (Fig. 1A). We characterized the somatic alterations in these cuproptosis regulators in each of the 33 tumors to better understand the landscape of the different tumor types. Different tumor types had different mutation patterns, and all 10 cuproptosis regulators exhibited mutations in bladder urothelial carcinoma (BLCA), colon adenocarcinoma (COAD), head and neck squamous cell carcinoma (HNSC), and uterine corpus endometrial carcinoma (UCEC), whereas no mutations were observed in diffuse large B-cell lymphoma, kidney chromophobe, testicular germ cell tumors, and uveal melanoma (UVM) (Fig. 1B). In addition, the highest mutation frequencies of CNKN2A were observed in HNSC (19.3%) and pancreatic adenocarcinoma (19.1%). For amplification, all cuproptosis regulators in BLCA, lung adenocarcinoma (LUAD), ovarian serous cystadenocarcinoma, and UCEC were amplified, whereas all those in UVM remained unchanged (Fig. 1C). DLD in esophageal carcinoma had the highest amplification frequency (8.5%). Consistent with Fig. 1A, a high percentage of CDKN2A deep deletion was observed in most tumor types (Fig. 1D). Among them, more than half of the glioblastoma multiforme samples showed deep deletion of CDKN2A, while UVM only showed a low percentage of deletions in FDX1 (1.3%) and DLAT (1.3%). Overall, cuproptosis regulators show a heterogeneous pattern of somatic alterations in different tumor types. Given that gene amplification and deep deletion largely mediate aberrant gene expression,28 we explored the effects of amplification and deep deletion on cuproptosis regulator expression in cancer. Consistent with expectations, amplification samples had the highest gene expression and deep deletion samples had the lowest gene expression among all 10 cuproptosis regulators (Supplementary information, Fig. S1). This suggests that SCNA affects the expression levels of cuproptosis regulators in tumors.
Gene expression patterns of cuproptosis regulators
To characterize the gene expression patterns of cuproptosis regulators, we first explored the interaction relationships among the regulators using the STRING database. As shown in Supplementary information, Fig. S2, the seven positive regulators and the negative regulator, GLS, formed an interaction network, whereas MTF1 and CDKN2A did not interact with other regulators. We further explored the distribution of regulator expression in different normal tissues. Overall, the expression of the regulators was evenly distributed in different tissues, with the highest expression of FDX1 in the adrenal gland and very low expression of CDKN2A in the bone marrow (Supplementary information, Fig. S3).
Differential expression analysis of paired normal and tumor tissues revealed that cuproptosis regulators were aberrantly expressed in 17 tumors (Fig. 2A). FDX1 expression was downregulated in 12 tumors, LIPT1 expression was downregulated in eight tumors, and CDKN2A expression was significantly upregulated in 16 tumors. Other regulators showed heterogeneous expression patterns; for example, LIAS expression was downregulated in most tumors, whereas its expression was increased in kidney chromophobe and LUAD. Subsequently, we explored the role of co-expression of the regulators in 33 tumors. The expression correlations prevalent among all regulators in all tumor types are shown in Supplementary information, Table S1. FDX1 expression was significantly positively correlated with the expression levels of six other positive regulators in most tumors. This suggests that cuproptosis regulators in tumors may co-regulate their cuproptosis activity.
To better understand the expression landscape of cuproptosis regulators in tumors, we performed unsupervised consensus clustering based on regulator mRNA expression for all samples of the 33 tumor types. Based on the consensus cumulative distribution function and delta area, all tumors were distinctly divided into two different sample clusters (Fig. 2B). Compared to cluster 2, cluster 1 had higher expression levels of FDX1, LIAS, LIPT1, DLD, DLAT, pyruvate dehydrogenase E1 subunit beta, and GLS, and significantly lower expression levels of CDKN2A (Fig. 2C). Examination of the distribution of all cancer types in the two clusters revealed that all acute myeloid leukemia, most COAD, three types of kidney cancers (kidney chromophobe, kidney renal clear cell carcinoma [KIRC], and kidney renal papillary cell carcinoma [KIRP]), prostate adenocarcinoma, rectum adenocarcinoma, testicular germ cell tumors, and thyroid carcinoma (THCA) were distributed in cluster 1, and most of the three types of gynecologic tumors (cervical squamous cell carcinoma and endocervical adenocarcinoma, ovarian serous cystadenocarcinoma, and uterine carcinosarcoma) were distributed in cluster 2 (Fig. 2D). As cuproptosis in tumors is regulated by 10 regulators, the expression of any individual regulator can hardly reflect the overall level of cuproptosis. Therefore, according to previous research methodology,29-31 we first proposed the cuproptosis positive score, cuproptosis negative score, and cuproptosis activity score based on the mRNA expression levels of positive and negative regulators of cuproptosis. In the pan-cancer context, the positive cuproptosis score was positively correlated with the expression of all positive regulators, and the negative cuproptosis score was positively correlated with the expression of all negative regulators (Supplementary information, Fig. S4). The cuproptosis activity score was positively correlated with the cuproptosis positive score as well as positive regulator expression and negatively correlated with the cuproptosis negative score as well as negative regulator expression (Supplementary information, Fig. S4); thus, the cuproptosis activity score integrated the expression abundance of all regulators and better reflected the overall cuproptosis level. Subsequently, we found that cluster 1 had significantly higher cuproptosis positive and activity scores, and lower negative scores than cluster 2 (Fig. 2E), suggesting that cluster 1 had a relatively higher level of cuproptosis. Survival analysis showed that cluster 1 had a significantly better overall prognosis than cluster 2 (Fig. 2F), implying that the cuproptosis level may influence the survival of patients with different tumor types.
Methylation analysis of cuproptosis regulators
Methylation in the promoter regions of genes largely regulates the gene expression, and hypermethylation generally suppresses the gene expression.32 However, there are some special cases where hypermethylation of the promoter region may enhance the expression of a gene, such as human telomerase reverse transcriptase.33 To explore the methylation alterations in cuproptosis regulators, we first compared the methylation differences in the regulators between paired normal and tumor tissues. As shown in Fig. 3A, the methylation levels of regulators varied in all 16 tumors. Methylation alterations in regulators were heterogeneous in different tumor types. For example, the methylation levels of FDX1 were elevated in BLCA, BRCA, COAD, HNSC, KIRP, and UCEC tumor tissues compared to paired normal tissues, while they were decreased in KIRC. Notably, MTF1 did not show altered methylation levels in any of the tumor types. Furthermore, we analyzed the correlation between regulatory methylation levels and mRNA expression levels in 33 tumor types. In most tumor types, the promoter methylation levels of FDX1, DLAT, GLS, and CDKN2A were negatively correlated with the mRNA expression levels (Fig. 3B). The methylation levels of LIAS, DLD, pyruvate dehydrogenase E1 subunit alpha 1, pyruvate dehydrogenase E1 subunit beta, and MTF1 were negatively or positively correlated with the mRNA expression levels in specific tumor types. For example, MTF1 was negatively correlated with BRCA, cervical squamous cell carcinoma and endocervical adenocarcinoma, DLBC, and brain low-grade glioma (LGG) and positively correlated with THCA (Fig. 3B). Notably, the methylation levels of LIPT1 were positively correlated to the mRNA expression levels in most tumor types. Survival analysis revealed that the methylation levels of cuproptosis regulators were correlated with the overall survival (OS) in 17 tumor types and that the correlations were tumor type-dependent. For instance, FDX1 hypermethylation was associated with poor OS in LGG and good OS in UVM (Fig. 3D, E).
MicroRNAs (miRNAs), long non-coding RNAs (lncRNAs), and transcription factors (TFs) regulate the expression levels of cuproptosis regulators
In previous results of this study, we described the regulation of cuproptosis regulators expression by SCNA and DNA methylation. miRNAs are gene expression repressors that can regulate gene expression post-transcriptionally by binding to the 3'-untranslated regions of target mRNAs.34,35 To comprehensively explore miRNAs that may regulate cuproptosis regulators, we screened all miRNAs that could target the 3'-untranslated regions of these regulators. In Supplementary information, Table S2, we have listed all potential miRNA–mRNA pairs that may target different cuproptosis regulators in each tumor type after threshold screening as a reference for future cuproptosis-related miRNA studies. Notably, 174 miRNA–mRNA pairs, including 127 miRNAs, were present in at least five tumor types. Interestingly, 33 of the 127 miRNAs targeted at least two cuproptosis regulators, constituting a miRNA regulatory network (Fig. 4A). Given that these miRNAs target multiple regulators in multiple tumors, these miRNAs may be potential miRNAs regulating cuproptosis.
LncRNAs are important regulators of gene expression and play important roles in transcription, translation and post-translational modifications.36 Therefore, we combined lncRNA regulation pan-cancer analysis with gene expression correlation analysis data to filter all potential lncRNA–mRNA pairs regulating cuproptosis in different tumor types (Supplementary information, Table S3).37 A total of 131 lncRNA–mRNA pairs were identified involving 55 lncRNAs, 34 of which targeted at least two cuproptosis regulators and constituted an lncRNA regulatory network (Fig. 4B).
TFs are key regulators of gene transcription and expression, and dysregulated TFs mediate aberrant gene expression and represent a unique class of drug targets.38 We examined a series of TFs listed in a previous pan-cancer study and identified a total of 465 TF–mRNA pairs containing 208 different TFs,37 of which 24 TFs regulated the expression levels of at least five cuproptosis regulators (Fig. 4C). Importantly, nuclear factor I C (NFIC) targeted all 10 cuproptosis regulators, suggesting that it may be a key TF mediating cuproptosis in tumors. Supplementary information, Table S4 lists all TF–mRNA pairs that may target cuproptosis regulators in different tumor types.
Cuproptosis activity predicts the prognosis of patients with cancer
The results of this study demonstrated that cuproptosis activity predicted the OS in a pan-cancer context. To further clarify the impact of cuproptosis on patient survival, we explored the prognostic predictive roles of cuproptosis regulators and cuproptosis scores for different tumor types. We performed Cox regression analysis on cuproptosis regulators or scores to calculate the survival risk and used the log-rank test to determine significance after dividing different tumors into two groups according to the median of cuproptosis regulators or scores. As shown in Fig. 5A, cuproptosis regulators and scores had different prognostic roles in different tumor types. High LIAS expression was associated with better OS in adrenocortical carcinoma, COAD, KIRC, KIRP, LGG, liver hepatocellular carcinoma (LIHC), and mesothelioma, while high CDKN2A expression was associated with poor OS in adrenocortical carcinoma, COAD, KIRC, LIHC, pheochromocytoma and paraganglioma, prostate adenocarcinoma, UCEC, and UVM. In addition, an elevated cuproptosis activity score was associated with poorer OS in HNSC and better OS in patients with KIRP, LIHC, and UCEC. Fig. 5B and C show the Kaplan–Meier survival curves between the high and low cuproptosis activity score groups after dividing the samples into two groups according to the median value in HNSC and UCEC, respectively. To further confirm the effect of cuproptosis on the OS of patients, we performed Kaplan–Meier survival curve analysis after dividing the patients into two groups according to the best cut-off value of the cuproptosis activity score. The results showed that the cuproptosis activity was significantly associated with the OS in 13 of the 33 tumor types (Supplementary information, Fig. S5). Among them, higher cuproptosis activity predicted better OS in COAD, diffuse large B-cell lymphoma, KIRP, acute myeloid leukemia, LGG, LIHC, lung squamous cell carcinoma, sarcoma, and UCEC and also predicted poorer OS in HNSC, LUAD, skin cutaneous melanoma, and UVM. In addition to OS, we also tested the relationship between cuproptosis and disease-free interval and progression-free interval in patients with tumors. We found that the cuproptosis regulator levels and activity scores were strongly correlated with disease-free interval in 16 tumor types and progression-free interval in 20 tumor types, with the correlation depending on the tumor type. These results suggest that cuproptosis is correlated with the prognosis of specific tumor types and can predict patient survival.
Pathway activity analyses of cuproptosis
Our results demonstrated the dysregulation and prognostic role of cuproptosis in tumors; however, the specific tumor-related pathways involved in cuproptosis remain unknown. Therefore, we first inferred the enrichment level of 50 cancer hallmark gene sets in all tumor samples from 33 tumor types (Supplementary information, Table S5), which comprehensively reflected the biological processes associated with tumors.39 Subsequently, we calculated the correlation of the cuproptosis activity score with all hallmark gene sets in each tumor type and generated a heatmap (Fig. 6A; Supplementary information, Table S6). Cuproptosis activity was positively correlated with approximately half of the hallmark gene sets in most tumor types and negatively correlated with the other half, indicating that cuproptosis plays a key role in tumors (Fig. 6A). Notably, cuproptosis activity was significantly positively correlated with oxidative phosphorylation in all 33 tumor types (Fig. 6B), which is consistent with the study by Tsvetkov et al. that cuproptosis is dependent on mitochondrial respiration and the tricarboxylic acid (TCA) cycle,26 further suggesting that the cuproptosis activity score can reliably reflect the cuproptosis level. In addition, cuproptosis was also significantly negatively correlated with hypoxia in 24 tumor types (Fig. 6B), and Tsvetkov et al. found that hypoxia (1% O2) reduces cuproptosis sensitivity by obliging cells to rely on glycolysis rather than oxidative phosphorylation.26,27 Cuproptosis was also significantly positively correlated to fatty acid metabolism in all tumor types (Fig. 6B). Additionally, cuproptosis was negatively associated with the apical junction, mitotic spindle, epithelial–mesenchymal transition (EMT), transforming growth factor (TGF)-β, KRAS, and tumor necrosis factor-alpha signaling pathways in more than 25 cancer types, confirming the important regulatory role of cuproptosis in tumor metastasis and growth. Strikingly, cuproptosis was also negatively associated with immune-related pathways, including the inflammatory response, complement, interleukin-6/Janus kinase/signal transducer and activator of transcription 3 signaling, and interferon gamma response pathways, in most tumor types (Fig. 6B). Given that cuproptosis was positively associated with DNA repair in 27 tumor types, we further explored whether cuproptosis was related to genomic instability. Consistent with expectations, cuproptosis was negatively associated with the indicators related to homologous recombination deficiency (HRD) in 11 tumor types, but negatively associated with mutational burden in only three tumor types (Supplementary information, Fig. S6). In summary, the above results suggest that cuproptosis is involved in numerous biological processes in tumors.
Cuproptosis-associated immune characteristics
In pathway activity analyses, we observed significant correlations between cuproptosis and immune-related pathways. To better reveal the intrinsic link between cuproptosis and tumor immunity, we first inferred the overall immune (ImmuneScore) and stromal (StromalScore) infiltration levels in all tumor samples (Supplementary information, Table S7). Consistent with the results of the pathway activity analyses, cuproptosis was significantly negatively correlated with the overall immune and stromal cell infiltration levels in most tumor types (Fig. 7A). In addition, cuproptosis was negatively correlated with the ESTIMATEScore of tumors, indicating that cuproptosis is positively correlated with tumor purity. To further understand the correlation between cuproptosis and the abundance of different types of immune cells, we collected the infiltration levels of 22 immune cells of all tumor samples from a previous pan-cancer analysis.40 We calculated the correlation between cuproptosis and the levels of immune cell infiltration in each tumor type (Supplementary information, Table S8). As shown in Fig. 7B, there was a heterogeneous correlation pattern between cuproptosis and tumor-infiltrating immune cells. In 11 tumor types, cuproptosis was negatively correlated with the abundance of M1 macrophages, whereas in 10 tumor types, cuproptosis was positively correlated with the abundance of follicular helper T cells (Fig. 7C). Overall, cuproptosis was correlated with the abundance of numerous immune cells, and the correlation may vary depending on the tumor type. Immunomodulators are crucial for the response to immunotherapy, and we collected some classical immune activation-related genes, immune checkpoint-related genes, and TGFβ/EMT pathway-related genes from a previous study by Zeng et al.41 We then calculated the correlation between cuproptosis and these immunomodulators (Supplementary information, Table S9). In general, cuproptosis was negatively correlated with most immunomodulators in most tumor types, which is consistent with the results of previous studies (Fig. 7D). Although immune activation-related genes appear to be positively associated with cuproptosis in some specific tumor types, such as glioblastoma multiforme, this correlation was not significant (Supplementary information, Table S9). Thorsson et al. classified all samples from 33 tumor types into six immune subtypes (C1-C6) with different immune characteristics in a pan-cancer context and were widely recognized,40 Therefore, we compared the differences in cuproptosis between the six immune subtypes. As shown in Fig. 7E, C4 (lymphocyte depleted) tumors had the highest level of cuproptosis activity, whereas C6 (TGF-β-dominant) tumors had the lowest cuproptosis activity. This further validated the negative correlation between cuproptosis and the TGF-β pathway. Based on this result, we can hypothesize that the lower immune response in high cuproptosis tumors may be related to lymphocyte depletion. These results imply a broad link between cuproptosis and tumor immunity.
Association of cuproptosis with drug sensitivity and immunotherapy outcome
After establishing the association of cuproptosis with numerous cancer hallmark pathways and immune-related characteristics, we aimed to understand whether cuproptosis could influence patient response to chemotherapy, targeted therapies, and immunotherapy. We integrated gene expression data and drug sensitivity data of cancer cell lines from the Genomics of Drug Sensitivity in Cancer database and analyzed the correlation of cuproptosis with the half maximal inhibitory concentration (IC50) of 198 drugs. Unexpectedly, we observed significant negative correlations between cuproptosis and the IC50 of 39 drugs but failed to observe a positive correlation between cuproptosis and the IC50 of any drug (Fig. 8A), suggesting that cuproptosis broadly increases the sensitivity of chemotherapy and targeted drugs. By examining the mechanism of action of these 39 drugs, it was found that some of the targeted pathways had been shown to be negatively associated with cuproptosis activity in the previous results. For example, cuproptosis in the pathway activity analyses was negatively correlated with the P53 and PI3K/MTOR pathways in 13 tumor types (Fig. 6B), and cuproptosis was associated with increased sensitivity to two P53 pathway inhibitors (MIRA-1 and Nutlin-3a (-)) and three PI3K/MTOR pathway inhibitors (Ipatasertib, LJI308, and Uprosertib) (Fig. 8A), which further enhanced the reliability of our study. Generally, a high cuproptosis activity indicates higher chemotherapy and targeted drug sensitivity.
Immune checkpoint inhibitor (ICI) therapy is currently the most successful and common immunotherapy approach.42,43 To further investigate whether cuproptosis affects ICI therapy outcomes in tumor patients, we collected a metastatic urothelial cancer cohort receiving anti-programmed death ligand-1 (PD-L1) therapy (IMvigor210) and a metastatic melanoma cohort receiving anti-programmed death-1 (PD-1) therapy (GSE78220) from previous studies.30,44 In the IMvigor210 cohort, cuproptosis was found to be negatively correlated with the tumor ImmuneScore, StromalScore, and ESIMATEScore (Fig. 8B). More importantly, cuproptosis was negatively correlated with PD-L1 expression in this cohort (Fig. 8C). Interestingly, Kaplan-Meier survival curve analysis showed that patients with high cuproptosis activity had longer survival after anti-PD-L1 therapy than those with low cuproptosis activity (Fig. 8D). In the GSE78220 cohort, although cuproptosis did not correlate with PD-1 expression (Fig. 8E), Kaplan–Meier survival curve analysis demonstrated that patients with high cuproptosis activity had a better OS after anti-PD-1 treatment than those with low cuproptosis activity (Fig. 8F). These results provide preliminary evidence of the predictive role of cuproptosis in ICI immunotherapy.