Multi-omics analysis uncovers clinical, immunological, and pharmacogenomic implications of cuproptosis in clear cell renal cell carcinoma

Transcriptome profiling (HTSeq-counts) of ccRCC (n = 539) and normal (n = 72) specimens together with clinical parameters were acquired from TCGA (https://​portal.​gdc.​cancer.​gov/​). Raw counts data were converted into transcripts per million (TPM), followed by log2 transformation. DNA methylation data (Methylation450K), somatic mutation (mutation annotation format), copy-number alteration (SCNA; Masked Copy Number Segment) and microRNA (miRNA) expression data of ccRCC patients were also collected from TCGA. Matrix files of transcriptome profiling and prognostic data of ccRCC patients (n = 101) were acquired from the E-MTAB-1980 dataset (https://​www.​ebi.​ac.​uk/​arrayexpress/​experiments/​E-MTAB-1980/​), which was utilized for external verification. The detailed clinicopathological information is listed in Additional file 2: Table S1.

Cuproptosis genes were collected from a previously published study [15]. Circos plot containing chromosome positions of cuproptosis genes was drawn via RCircos package [12].

For DNA methylation data, only the probes that mapped to the promoter region of cuproptosis genes were retained. For genes with multiple probes, the average β value of all probes was utilized as the methylation level.

To lower the false-positive rate, only non-silent mutations were retained. Through maftools package [16], the mutation annotation format from TCGA was analyzed. For SCNA, significant amplifications and deletions were identified via GISTIC2.0 [17]. Then, OncoPrint plots of mutations and SCNA were produced with ComplexHeatmap package [18].

The enrichment score of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets was evaluated between low and high expression of CDKN2A groups using gene set enrichment analysis (GSEA) [19]. GO and KEGG enrichment analysis of cuproptosis-related genes was conducted via clusterProfiler package [20]. Terms with false discovery rate (FDR) < 0.05 were regarded as significant enrichment. Gene set variation analysis (GSVA) [21], and fast GSEA (fGSEA) were used to compute and compare the activity of the 50 hallmark pathways based on the Molecular Signatures Database-derived “h.all.v7.4.entrez.gmt” gene set as the reference [22].

ccRCC samples were classified as distinct cuproptosis subtypes based on the transcriptome profiling of cuproptosis genes using unsupervised clustering algorithm. The number of clusters was identified via ConsensusClusterPlus package [23], with iterations for increasing the classification reliability.

The top 200 up-regulated genes in each subtype versus others were identified via limma package [24], which were used to validate the cuproptosis subtypes through nearest template prediction (NTP) algorithm in the E-MTAB-1980 cohort [25].

Single-nucleotide variant (SNV) neoantigens, tumor mutation burden (TMB), SCNA, cancer-testis antigen (CTA) score, homologous recombination defects, intratumor heterogeneity, and aneuploidy score were acquired from previously published literature [26]. In addition, mRNA expression-based stemness index (mRNAsi) was computed utilizing one-class logistic regression machine-learning approach [27].

Drug sensitivity data of cancer cell lines (CCLs) were acquired from Genomics of Drug Sensitivity in Cancer (GDSC) [28], Cancer Therapeutics Response Portal (CTRP) [46], and PRISM [29] databases. GDSC covers the half-maximal inhibitory concentration (IC50) data; meanwhile, both CTRP and PRISM cover the area under the curve (AUC) data as an evaluation indicator of drug sensitivity. Transcriptome profiling of CCLs was acquired from the Cancer Cell Line Encyclopedia (CCLE) [30]. The IC50 of GDSC-derived compounds was evaluated through oncoPredict package [31]. CTRP- and PRISM-derived compounds with more than 20% missing data were removed before K-nearest neighbor imputation. Next, pRRophetic package was employed to estimate the AUC of compounds [32]. Spearman correlation analysis on Cuscore and IC50 or AUC was executed to evaluate the drug response of ccRCC patients to each compound.

Single-sample GSEA (ssGSEA) was utilized to quantify the relative abundance of 22 tumor-infiltrating immune cell types and two stromal components (fibroblasts and endothelial cells) based on the known marker genes (Additional file 3: Table S2) [33]. Immune score and stromal score were estimated through ESTIMATE algorithm [34].

Multi-omics profiling (comprising mRNA expression, SCNA, and DNA methylation) of 75 immunomodulators was observed across distinct cuproptosis subtypes (Additional file 4: Table S3) [35].

Subclass Mapping (SubMap) analysis was adopted to assess the expression similarity between the three cuproptosis subtypes and immunotherapeutic responses [36]. The degree of commonality between the two groups was inferred via GSEA algorithm. Bonferroni-corrected p < 0.05 denoted that two groups were significantly similar.

To identify the cuproptosis-related genes based on the three cuproptosis subtypes, differentially expressed genes (DEGs) were screened through limma package with the threshold of |fold change|> 1.5 together with adjusted p < 0.05. Prognostic value of cuproptosis-related genes was assessed via univariate Cox regression analysis. Then, genes with p < 0.05 were retained for computing the Cuscore through adopting PCA, with the principal components (PCs) 1 and 2 as the final signature score. The Cuscore was defined as Cuscore = \(\sum_{n}^{m}(PCn+PCm)\), where n and m denoted the order and total number of prognostic cuproptosis-related genes in ccRCC, respectively.

Univariate Cox regression on clinicopathological parameters together with Cuscore was analyzed across ccRCC patients. Thereafter, significant prognostic predictors (p < 0.05) were retained for multivariate Cox regression analysis, and a nomogram was defined through adopting independent predictive variables (p < 0.05). The consistency between predicted and actual prognostic outcome was assessed via calibration curves. Through decision curve analysis, the net benefits of the nomogram, Cuscore and other clinicopathological parameters were measured.

Figure 2A illustrates the genomic location of each cuproptosis gene. Most cuproptosis genes (FDX1, DLD, DLAT, PDHA1, PDHB, and GLS) were notably down-regulated in ccRCC versus normal tissues, with only up-regulated CDKN2A (Fig. 2B, C). At the transcriptional level, cuproptosis genes positively interacted except CDKN2A in ccRCC (Fig. 2D). Next, we analyzed the reasons for the low expression of cuproptosis genes. Higher methylation levels of most cuproptosis genes were observed in ccRCC than normal tissues (Fig. 2E). In addition, copy number deletion of cuproptosis genes occurred in ccRCC, especially PDHB (Fig. 2F). Hypermethylation and copy number deletion might contribute to the low expression of cuproptosis genes. Their prognostic value was then assessed. Consequently, most cuproptosis genes were linked with better OS outcome of ccRCC, with opposite effect of CDKN2A (Fig. 2G). PCA revealed the remarkable discrimination of transcriptome profiling of cuproptosis genes between ccRCC and normal samples (Fig. 2H). Altogether, cuproptosis might be indispensable for ccRCC initiation and progression. ccRCC is highly immune infiltrated; adaptive and innate immune cells infiltrate the tumor microenvironment and constitute an ecosystem that regulates each aspect of ccRCC development [37]. Most cuproptosis genes were negatively correlated to the infiltration of immune cells, with positive relationships between CDKN2A and immune cell infiltration (Fig. 2I). Based on above findings, CDKN2A was identified as a cuproptosis gene of interest. CDKN2A expression was higher in more advanced grade and stage (Fig. 2J, K). GSEA revealed that CDKN2A positively correlated to immunity (activation of immune response, immunoglobulin production, lymphocyte-mediated immunity, etc.) (Fig. 2L).

According to the transcriptome profiling of cuproptosis genes, cuproptosis subtypes were established via unsupervised clustering analysis. When the consensus value k was 3, most of the colors in the consensus matrix did not overlap (Fig. 3A). Cumulative distribution function (CDF) as well as item tracking demonstrated the cluster stability when k = 3 (Additional file 1: Figure S1A–C). Taken together, we classified ccRCC as three cuproptosis subtypes, namely C1 (moderate cuproptosis), C2 (low cuproptosis), and C3 (high cuproptosis) (Additional file 5: Table S4; Fig. 3B). PCA illustrates the notable discrepancy of transcriptome profiling among the three cuproptosis subtypes (Fig. 3C). Prognosis difference was observed among the three cuproptosis subtypes, with the best OS in C3, intermediate in C1, and worst in C2 (Fig. 3D). According to fold change > 1 together with adjusted p < 0.05, we determined up-regulated genes in each cuproptosis subtype (Fig. 3E). The top 200 genes were regarded as up-regulated biomarkers of each cuproptosis subtype (Additional file 6: Table S5). To further verify the stability and robustness of cuproptosis subtypes, NTP analysis that may quantify the prediction confidence of each patient based on transcriptome profiling was implemented in the E-MATB-1980 cohort. Consequently, the three cuproptosis subtypes exhibited high reproducibility in the E-MATB-1980 dataset (Fig. 3F). The differences in OS outcome and transcriptome profiling among the three cuproptosis subtypes were proven in this cohort (Fig. 3G, H). Clinicopathological features were prominently different among the three cuproptosis subtypes, with the lowest proportions of grade and stage in C3, intermediate in C1, and highest in C2 (Fig. 3I). Altogether, the cuproptosis-based classification was reproducible and stable in ccRCC.

DNA alterations mainly comprise somatic mutation and SCNA. Overall, somatic mutation exhibited the remarkable heterogeneity among the three cuproptosis subtypes (Fig. 4A–C). Of note, C3 had the highest mutation frequency of VHL (61%), followed by C1 (57%) and C2 (54%). PBRM1 had the highest mutation frequency in C1 (41%), intermediate in C3 (33%), and lowest in C2 (20%). Overall, C3 presented significantly lower SNV neoantigens and TMB score relative to C1 (Fig. 4D, E). Next, we observed the heterogeneity of SCNA profiling in the three cuproptosis subtypes (Fig. 4F). Additionally, C1 presented higher SCNA, CTA score, homologous recombination defects, and intratumor heterogeneity (Fig. 4G–J). We computed mRNAsi for quantifying cancer stemness, and found that C3 had higher mRNAsi (Fig. 4K). Drug sensitivity differences were also assessed among the three cuproptosis subtypes. Consequently, C1 patients had the highest sensitivity to pazopanib, and sorafenib, while C2 patients were most sensitive to sunitinib (Fig. 4L–N).

Next, we observed the remarkable differences in the oncogenic hallmark pathways among the three cuproptosis subtypes, contributing to the intratumor heterogeneity of ccRCC (Fig. 5A). C1 subtype presented the highest expression of common immune checkpoint molecules (PDCD1, CTLA4, etc.) (Fig. 5B). Additionally, the most abundant infiltration of immune cells was observed in C2, with intermediate for C1, and lowest for C3. In Fig. 5C, there were extensive genomic differences in immunomodulators across the three cuproptosis subtypes. Of note, C3 had the highest frequencies of amplification (IFNG, CD27, LAG3, CD40, GZMA, etc.) and deletion (PDCD1LG2, CD274, BTN3A1, BTN3A2, TNF, VEGFA, IFNA1, IFNA2, MICA, MICB, TLR4, PRF1, ENTPD1, etc.) of several immunomodulators, followed by C1 and C2. Considering that the cuproptosis subtypes appear to correlate to the tumor immune microenvironment, we inferred the immunotherapeutic responses of the three cuproptosis subtypes. Consequently, C1 subtype exhibited the high expression similarity to response to anti-PD-1 therapy (Fig. 5D), proven in the E-MTAB-1980 dataset (Fig. 5E). Hence, patients in C1 subtype clinically benefited more from anti-PD-1 immunotherapy.

To assess the functional role of the three cuproptosis subtypes, 95 cuproptosis-related genes were determined (Fig. 6A; Additional file 7: Table S6) through intersecting DEGs between cuproptosis subtypes according to |fold change|> 1.5 and adjusted p < 0.05. Functional enrichment analysis revealed that they were remarkably enriched in diverse metabolism processes (Fig. 6B). Next, the prognostic value of cuproptosis-related genes was investigated. Consequently, all of them were significantly linked with ccRCC patients’ OS (Additional file 8: Table S7). Through PCA computational approach, the Cuscore was defined based on cuproptosis-related genes to quantify cuproptosis of individual ccRCC patients. Next, we investigated the clinical relevance of the Cuscore. ccRCC patients were classified as low and high Cuscore groups, with the prominent distinction of transcriptome profiling (Fig. 6C). Patients with high Cuscore possessed a remarkable survival benefit (Fig. 6D), with 1-, 3- and 5-year OS AUC values of 0.74, 0.69 and 0.70, respectively (Fig. 6E). The E-MTAB-1980 dataset proved the high reproducibility of this Cuscore (Fig. 6F–H).

Next, clinicopathological characteristics of low and high Cuscore groups were observed. In Fig. 7A, low Cuscore patients exhibited higher proportions of dead status, male, advanced stage, and grade. Additionally, male patients had lower Cuscore relative to female patients (Fig. 7B). With the increase in grade and stage, the Cuscore gradually lowered (Fig. 7C, D). Altogether, the Cuscore may reflect tumor progression and prognostic outcome.

From uni- and multivariate Cox regression results, Cuscore served as an independent protective factor, with clinicopathological variables (age, stage, and grade) as independent risk factors (Fig. 7E, F). To facilitate the clinical utility of the Cuscore in predicting the survival probability, we generated a nomogram that integrated above independent prognostic factors. Calibration curves of 1-, 3- and 5-year OS displayed that the nomogram was close to the actual survival probability (Fig. 7G–I). Moreover, we found that the nomogram had the highest net benefit in clinical assessment based on decision curve analysis, suggesting the significance of the Cuscore in coordination with other clinicopathological variables for survival prediction (Fig. 7J–L).

To understand the immunogenomic features involved in the Cuscore, we evaluated the relationships of the Cuscore with immunogenomic signatures. Consequently, the Cuscore was negatively linked with aneuploidy score, CTA score, homologous recombination defects, and intratumor heterogeneity in ccRCC (Fig. 8A–D). Additionally, we observed the negative associations of the Cuscore with most immune cells (Fig. 8E), and most immune checkpoint molecules were up-regulated as the Cuscore increased (Fig. 8F). The fGSEA revealed that the Cuscore was prominently correlated to oncogenic hallmark pathways (Fig. 8G).

Next, this study analyzed miRNAs with differential expression between low and high Cuscore groups. Sixty-nine miRNAs that presented significant differential expression were selected (Additional file 9: Table S8), and KEGG pathway enrichment analysis of their targeted mRNAs was implemented. Consequently, pathways in cancer, cell adhesion molecules (CAMs), Hippo, FoxO and TGF-β signaling pathways were notably enriched (Fig. 8H). Most miRNA-targeted mRNAs in above pathways were up-regulated in high Cuscore group. Above evidence suggested that Cuscore was linked with post-transcriptional mechanisms and pathway regulation.

We further understood the effects of the Cuscore on drug response. The associations of the Cuscore with the response to GDSC-derived compounds were then assessed. The sensitivity to nine drugs (AGI-6780, topotecan, gefitinib, erlotinib, sapitinib, ibrutinib, Eg5_9814, palbociclib and KRAS (G12C) Inhibitor-12) correlated to the high Cuscore, with the correlation between P22077 resistance and the high Cuscore (Fig. 9A). Next, we investigated the pathways involved in these drugs. As illustrated in Fig. 9B, drugs whose sensitivity was linked with the high Cuscore were mainly targeting cell cycle, DNA replication, EGFR and ERK MAPK signaling. In addition, the CTRP and PRISM databases were utilized to predict potential compounds for ccRCC. CR-1-31B, leptomycin B, paclitaxel, vincristine, ouabain, BI-2536, methotrexate, combretastatin-A-4, cabazitaxel, vincristine, PHA-793887, romidepsin, dolastatin-10, gemcitabine, and YM-155 were more suitable for patients with low Cuscore (Fig. 9C, D). Altogether, our findings implied that cuproptosis was correlated to drug response. Therefore, the Cuscore might be a potential biomarker for formulating appropriate therapeutic schedule.

Potentially druggable targets for patients with low Cuscore were identified. Firstly, correlation between the Cuscore and protein expression of druggable targets was computed, and 23 protein targets were selected according to correlation coefficient < −0.5 and p < 0.05 (Fig. 9E). Then, we conducted correlation analysis of CERES score with the Cuscore, and further screened 486 targets based on correlation coefficient > 0.75 and p < 0.05 (Fig. 9F). Two genes (NMU, RARRES1) were finally determined as potential therapeutic targets by above methods.