The signature based on CRLM could well predict the survival of ccRCC patients.
A total of 245 cuproptosis-related lncRNAs were obtained by Pearson's test and displayed by the Sankey plot in Fig. 2a. The most prognosis related lncRNAs and mRNAs were assessed by univariable COX regression analysis (Fig. 2b). The lncRNAs or mRNAs which hazard ratio > 1 were defined as risk factors, including LINC01711, SNHG3, LINC00460, MANCR, and AL133215.2. On the contrary, AL731577.2, DLAT and DBT were considered as protect factors. To avoid the influence of confounding factors, we performed a LASSO regression analysis to re-assess the 8-prognosis related CRLM and 4 of them with nonzero coefficients (Fig. 2c, 2d). After evaluated with AIC value, we retained 4 CRLM to build the prediction signature including AL731577.2, LINC00460, AL133215.2, and DBT (Table 1).
ccRCC samples were then randomly divided into training set (N = 265) and testing set (N = 265). The Chi-square test revealed that there was no significant difference in clinical features between training and testing sets (Table S2). Hence, in addition to the risk scores, other risk factors can be well excluded. Kaplan-Meier analysis indicated that patients in high-risk group suffered from worse prognosis than that in low-risk group (Fig. 3a). The distribution of risk scores, the survival outcomes of patients in different groups, and the expression profiles of the 4 CRLM are shown in Fig. 3b. Along with the increase in risk score, the survival status of the three groups showed basically the same trend and more dead cases accumulated. Besides, the expression of AL731577.2, DBT decreased and LINC00460, AL133215.2 increased as the risk score increased.
These results suggested that the signature based on CRLM could well predict the survival of ccRCC patients.
The signature showed well prognostic value for ccRCC patients.
The univariable and multivariable Cox regression analyses revealed that the signature of 4 CRLM was an independent prognostic factor for ccRCC patients. The univariable Cox regression showed that the HR was 1.074 and 95% CI was 1.034–1.117 (Fig. 4a, P < 0.001). The multivariable Cox regression showed that the HR was 1.047 and 95% CI was 1.022–1.073 (Fig. 4b, P < 0.001). The AUC of the risk score (AUC = 0.731) was higher than the AUCs of other clinicopathological features including age (0.633), gender (0.511), and grade (0.722), indicating that the prognostic signature for ccRCC was reasonably dependable (Fig. 4c). The AUC of the novel CRLM signature for 1-, 3-, 5-year survival rates were 0.731, 0.714, 0.739, respectively (Fig. 4d). In addition, we compared the CRLM signature with 4 classifiers which have been reported to be feasible for predicting ccRCC patients’ outcomes. The AUCs of CRLM signature were higher than Ferroptosis signature (18) (Fig. 4e, PMID: 34281531), Lindsay signature (17) (Fig. 4f, PMID: 34670568), Joanna signature (16)(Fig. 4g, PMID: 26631499), and Qu signature (15) (Fig. 4h, PMID: 30143382) for 1-, 3-, 5-year survival rates. Furthermore, with the extension of survival time, the C-index of the risk score gradually approached that of the tumor stage (Fig. 4i). Among the above 5 signatures, the CRLM signature achieved the highest C-index (0.692, Fig. 4j).
We then established an accurate and stable hybrid nomogram that contains clinic-pathological characteristics (age, gender, stage, and grade) and the novel CRLM signature (Fig. 4k). To use the nomogram, the specific points (black dots) of individual patients are located on each variable axis. Red lines and dots are drawn upward to determine the points received by each variable; the sum (198) of these points is located on the Total Points axis, and a line is drawn downward to the survival axes to determine the probability of 5- year (84.4%) ,3-year (90.6%), and 1-year (96.7%) overall survival. In this combined nomogram, the risk score model was proven to exert excellent weight among all these clinically relevant covariates, which was similar to the findings from the multivariable Cox regression analysis. Thus, the CRLM signature may be applied in the clinical management of ccRCC patients. The calibration curve for the overall survival nomogram model demonstrated good agreement between prediction and observation (Fig. 4l). The Decision Curve Analysis (DCA) disclosed a better net benefit of nomogram than risk and grade (Fig. 4m).
The CRLM signature also displayed a good performance in the Japanese ccRCC cohort. The association between the risk group and clinicopathological characteristics (gender, age, grade, stage) of patients was summarized in Table 2. The results showed that high risk group was significantly associated with advanced tumor stage (p = 0.001). Besides, high risk group patients were estimated lower survival (Fig. 5a). The AUC of risk score is 0.868 (Fig. 5b), which is higher than stage (0.815) and grade (0.765). The C-index of risk score was higher than stage when predicting survival time over 5 years (Fig. 5c). With the increasing risk score, more death cases with lower survival time were concentrated in the high-risk group. Moreover, the expression profiles of the 4 CRLM in the Japanese ccRCC cohort were consistent with the TCGA-KIRC cohort basically (Fig. 5d).
Table 2
Correlations between the risk score and clinicopathological features in 98 Japan ccRCC patients.
Covariates
|
Type
|
Total
|
low-risk
|
high-risk
|
p-value
|
Age
|
<=65
|
56
|
30
|
26
|
0.726
|
|
> 65
|
42
|
21
|
21
|
|
Gender
|
FEMALE
|
23
|
13
|
10
|
0.623
|
|
MALE
|
75
|
38
|
37
|
|
Grade
|
G1-G2
|
71
|
39
|
32
|
0.353
|
|
G3-G4
|
27
|
12
|
15
|
|
Stage
|
Stage I-II
|
73
|
45
|
28
|
0.001
|
|
Stage III-IV
|
25
|
6
|
19
|
|
Principal‑component analysis (PCA) further verifies the grouping ability of the CRLM signature.
PCA was performed to prove the difference between high and low-risk groups based on the entire gene expression profiles, 16 cuproptosis -related mRNAs, 245 cuproptosis-related lncRNAs, and the signature classified by the expression profiles of the 4 CRLM.
Figure 6a–c shows that the distribution of high and low-risk groups was relatively dispersed. However, the results derived from our signature indicate that high-risk and low-risk groups have different distributions (Fig. 6d). These results revealed that the prognostic characteristics can distinguish between high-risk and low-risk groups.
Gene Mutation Analysis
We processed simple nucleotide variation data using the “maftools” package in R software. A waterfall plot displayed the top 15 mutated genes in patients with ccRCC (Fig. 7a, 7b). A higher proportion of somatic mutations (SETD2) was found in the high-risk group. Nevertheless, there was no difference between the high and low-risk groups in the TMB (P = 0.068, Fig. 7c). High TMB patients had a lower survival rate than that with the low TMB (Fig. 7d, p = 0.001). The high TMB + high risk group patients suffered from the poorest survival time than the other groups. On the contrary, the low TMB + low risk group patients were predicted with longer survival time significantly (Fig. 7e, p < 0.001).
Gene set enrichment analysis for CRLM signature.
The GSEA analysis indicated that the Gene Ontology correlated with the high-risk group, which may be involved in the regulation of immune system (Fig. 8a). The humoral immune response in Biological Process, the immunoglobulin complex in Cellular Component, and immunoglobulin receptor binding in Molecular Function were all significantly enriched in high-risk group. The KEGG pathways further disclosed the close correlation between the high-risk group and the regulation pathway of cytokine-cytokine receptor signaling (Fig. 8b). These results drove us to explore the difference between high and low risk groups in immune function and immune therapy response.
The heatmap of immune functions based on ssGSEA of TCGA-KIRC data revealed that most immune functions were significantly different between high-risk and low-risk groups, except for MHC class Ⅰ, HLA, and APC-co-inhibition (Fig. 8c). Differing from the other immune functions, type Ⅱ IFN response was shown to be inhibited in the high-risk group.
Validation of the immune therapy and targeted therapies response with the CRLM signature.
ICIs and targeted therapies are usually applied as the first-line treatment strategy for advanced ccRCC patients after surgery. Hence, we first divided TCGA-KIRC cohort into stage Ⅰ-Ⅱ and stage Ⅲ-Ⅳ groups. Kaplan–Meier survival analysis was performed subsequently (Fig. 9a, 9b). The results revealed that the risk type can better distinguish the survival curves of advanced patients (stage Ⅲ-Ⅳ).
We then used TIDE to assess the potential clinical efficacy of immunotherapy for advanced samples (Fig. 9c). A higher TIDE prediction score represents a higher potential for immune evasion, suggesting that patients are less likely to benefit from ICI treatment. The results indicated that the high-risk group had a higher TIDE score than the low-risk group, implying that patients in the low-risk group could benefit more from ICI treatment than patients in the high-risk group. In addition to ICI treatment, we attempted to identify the association between the signature and the efficacy of 3 tyrosine kinase inhibitor (TKI) drugs which were commonly used in clinic for the treatment of ccRCC. We found that the low-risk group was associated with lower IC50 for Pazopanib (Fig. 9d, p < 1.8e-12) and Sunitinib (Fig. 9e, p < 2.3e-08). On the contrary, the low-risk group was associated with higher IC50 for Sorafenib (Fig. 9f, p < 4.6e-09). These results suggested that high-risk group patients may be sensitive to Sorafenib, and low-risk group patients may be sensitive to Sunitinib and Pazopanib. Hence, we proposed that the CRLM signature could be used as a potential predictor for targeted therapies sensitivity.