The work flow of this study is shown in Fig. 1. First, a total of 260 stage III/IV ovarian cancer patients with platinum treatment from GSE32062 were enrolled to construct a Cox model to predict the prognosis. Second, a total of 185 and 110 stage III/IV ovarian cancer patients with platinum treatment respectively from GSE26712 and GSE17260 were used as the validation group (validation group 1 and 2) to verify the models. The basic clinical characteristics of enrolled patients are listed in Table 1.
Identification of the prognostic ferroptosis-related and necroptosis-related genes in the GSE32062 cohort
A total of 85 ferroptosis-related genes and a total of 159 necroptosis-related genes were used to analyze the relationship with prognosis of ovarian cancer patients in the GSE32062 cohort, respectively. Based on the expression profile, through the univariate and multivariate Cox regression analyses with OS, eight DEGs related to ferroptosis were finally identified closely related to OS, including NFS1, ATG7, G6PD, VDAC2, SLC3A2, MAP1LC3C, ACSL3, and PTGS2 (Fig. 2a). Through the same method, we screened 10 necroptosis-related DEGs associated with OS of ovarian cancer patients, namely STAT5B, CAMK2D, HIST1H2AJ, CASP1, PYGB, IFNAR2, CAMK2G, STAT1, FADD, and HMGB1 (Fig. 2b). The functions and ovarian cancer-related research of these genes are listed in Supplementary Table S1.
Construction of four ferroptosis-related and necroptosis-related prognostic models in the GSE32062 cohort
Based on the identified DEGs, we constructed two prognostic models for ovarian cancer through Cox regression analyses. The risk score was calculated by the DEGs (n(ferroptosis) = 8, n(necroptosis) = 10), and the Yoden index was used to calculate the best cutoff value to divide the enrolled ovarian cancer patients into high and low risk groups (Fig. 2c, d). The characteristics of two models were described as follows.
Ferroptosis-related prognostic model
We used eight screened DEGs to construct the ferroptosis-related prognostic model with OS. The OS of ovarian cancer patients prolonged with the high-expression of ATG7, G6PD, SLC3A2, MAP1LC3C and PTGS2, but shrank with the increased expression of NFS1, VDAC2, ACSL3 (Fig. 2a). According to the best cutoff value of risk score, we divided 260 enrolled ovarian cancer patients into high-risk group (n = 116) and low-risk group (n = 144), shown in Fig. 2c. Survival analysis showed that the OS of high-risk group was significantly shorter than low-risk group (P < 0.0001; Fig. 3a). The risk score for OS was calculated by the predictive performance of the eight-gene receiver operating characteristic (ROC) curves. The area under the curve (AUC) reached 0.632 of 3-year survival, 0.683 of 5-year survival, and 0.681 of 10-year survival (Fig. 3b).
Necroptosis-related prognostic model
To build the necroptosis-related OS prognostic model, we used ten selected DEGs of STAT5B, CAMK2D, HIST1H2AJ, CASP1, PYGB, IFNAR2, CAMK2G, STAT1, FADD and HMGB1.The DEGs of STAT5B, CAMK2D, HIST1H2AJ, IFNAR2, STAT1 and FADD were identified as six protective factors to this model, and CASP1, PYGB, CAMK2G, and HMGB1 were identified as four risk factors (Fig. 2b). Survival analysis also showed a low-risk group was much longer than high-risk group in OS (P < 0.0001; Fig. 3c). The AUC of the ten-gene ROC curves reached 0.660 of 3-year survival, 0.728 of 5-year survival, and 0.719 of 10-year survival (Fig. 3d).
Validation of the prognostic models in the GSE26712 cohort
Here, we used GSE26712 cohort to validate the ferroptosis-related and necroptosis-related prognostic models with OS. The AUC of ferroptosis-related ROC was 0.584 of 3-year survival, 0.600 of 5-year survival, 0.663 of 10-year survival (Fig. 4b). And the AUC of necroptosis-related ROC was 0.634 of 3-year survival, 0.624 of 5-year survival, 0.580 of 10-year survival (Fig. 4d).
Following the same formula of risk score from GSE32062 cohort, we calculated risk score of each patient enrolled from the GSE26712 cohort. According to the risk scores, the GSE26712 cohort were divided into high-risk group and low-risk group. The OS of high-risk group was significantly shorter than low-risk group in ferroptosis-related prognostic model (P = 0.0085; Fig. 4a), and the similar result was also shown in necroptosis-related prognostic model (P = 0.0049; Fig. 4c). Therefore, the ferroptosis- and the necroptosis-related prognostic models with OS for ovarian cancer also worked well in the GSE26712 cohort.
Validation of the prognostic models in the GSE17260 cohort
To ensure the robustness of the two models in our study, a total of 110 ovarian cancer patients from the GSE17260 cohort were enrolled for further validation. The patients were clustered into high-risk group and low-risk group following the same way used in the GSE32062 cohort. The AUC of 3-, 5- and 10-year survival were 0.573, 0.588 and 0.594 in ferroptosis-related ROC (Fig. 5b), and 0.559, 0.595, 0.610 respectively in necroptosis-related ROC (Fig. 5d). Meanwhile, the survival curve showed significant differences between high and low risk groups in ferroptosis-related model (P = 0.015; Fig. 5a), but no statistical significance in the necroptosis-related model (P = 0.072; Fig. 5c). The results revealed that the ferroptosis-related prognostic model with OS for ovarian cancer worked well in the GSE17260 cohort, but the necroptosis-related model should be further optimized in the future.
Functional enrichment analysis in the GSE32062 cohort
To explore the biological functions and regulatory pathways related to the prognostic models, we selected the DEGs between the high-risk group and low-risk group to conduct GO and KEGG analyses. Interestingly, many immune related functions and pathways were enriched in both ferroptosis-related and necroptosis-related risk models (Fig. 6a, d).
In the ferroptosis-related prognostic model, 10 immune-related GO terms were significantly enriched in (P < 0.05), including five biological processes (BPs) of immune response, humoral immune response, innate immune response, chemokine-mediated signaling pathway, positive regulation of NF-κB transcription factor activity; and five immune-related molecular functions (MFs) of chemokine activity, cytokine activity, Toll-like receptor 4 binding, CCR chemokine receptor binding and CXCR chemokine receptor binding (Fig. 6b). Besides, six immune-related KEGG pathways were also enriched in the ferroptosis-related prognostic model, including cytokine-cytokine receptor interaction, chemokine signaling pathway, TNF signaling pathway, NF-kappa B signaling pathway, NOD-like receptor signaling pathway and Toll-like receptor signaling pathway (Fig. 6c).
In the necroptosis-related prognostic model, two immune-related BPs were enriched, including chemokine-mediated signaling pathway and immune response; three immune-related MFs were enriched, including chemokine activity, CCR chemokine receptor binding and CXCR3 chemokine receptor binding; two immune-related KEGG pathways were enriched, including chemokine signaling pathway and cytokine-cytokine receptor interaction (Fig. 6e, f).
Microenvironment analysis in the GSE32062 cohort
According to the functional enrichment analysis, immune-related factors were identified as key elements to ovarian cancer patients in the GSE32062 cohort. To comprehensively analyze the importance of the immune-related factors in the two above prognostic models, 45 immune-related indexes calculated from the gene expression data were included.
In the ferroptosis-related prognostic model, 43 of the 45 immune-related factors had significant relationship with risk score, and plasma cell and mDC were the two exceptions. As expected, almost all the immune-related factors had negative relationships with risk scores, except activated CD4 and activated CD8 (Fig. 7a). For the 10 DEGs enrolled in the ferroptosis-related prognostic model, we comprehensively analyzed the correlation between each DEG expression and immune-related factors. NFS1, ATG7, VDAC2 and PTGS2 were four genes associated with more than five factors. NFS1 and VDAC2 were two DEGs that were negatively associated with immune-related factors, involving 9 and 5 factors, respectively. ATG7 and PTGS2 were positively associated with 13 and 21 immune-related factors, respectively (Fig. 7b).
In the necroptosis-related prognostic model, 35 of 45 immune-related factors had significant relationship with risk score, and all of them showed positive correlation (P < 0.05; Fig. 7a). For the 12 DEGs enrolled in this model, except CAMK2D-NK56 dim and IFMAR2-activated CD8, all the significant relationships between each gene and factors were positive. For the correlation analysis of each DEG with immune-related factors, five DEGs were associated with more than 30 immune-related factors, including CASP1, IFNAR2, STAT1, CYLD and STAT2. Among them, CASP1 was a DEG most associated with immune-related factors, involving 40 factors. On the contrary, five DEGs were associated with less than 2 immune-related factors, including HIST1H2AJ, PYGB, CAMK2G, FADD and HMGB1. CAMK2G was a DEG associated with none of the 45 immune-related factors (Fig. 7c).